According to EpochAI’s Capability Index, Chinese AI models have historically lagged the US frontier capabilities by 7 months, on average, over the past 3 years.

With the recent release of Moonshot AI’s Kimi K3, that gap is now estimated to be closer to three months.

On July 16, 2026, Moonshot AI, a leading Chinese AI lab, launched Kimi K3, a 2.8T parameter mixture-of-experts (MoE) model awarding a #3 spot on the Artificial Analysis’s Intelligence Index, behind only Claude Fable 5 and GPT-5.6 Sol.

While the American frontier models are still setting the pace, they are doing so with closed-weight, proprietary models - even though American technology companies compete in both categories. The Chinese labs have distributed their models almost entirely as open-weight.

According to Stanford’s HAI 2026 Index, “the performance gap between leading closed-weight and open-weight models have fluctuated over the past three years, with open-weight systems closing in and then falling behind as new proprietary models are released. While closed-weight models still lead, open-weight models are far more competitive than they were a few years ago.

The United States’ substantial lead in 2023 shrank considerably by early 2025, and the performance gap has remained narrow since then. Over the past year, the gap has fluctuated between near parity and low single digits. This convergence is particularly notable because it has emerged from two distinct development environments and institutional contexts.”

What happens if the world builds on Chinese open models? That is what podcast pontificators, national security experts, and technologists are asking.

My read is that China is striving for geopolitical dominance. If developers around the world standardize on Chinese open-weight models, the Chinese could gain influence over the foundational layer of the AI ecosystem much like Android became a default foundation that millions of other products were built on.

Build on the Chinese models, they are free!

Free, sorta. The biggest strategic issue isn’t whether the weights are open or closed - it’s who owns the surrounding ecosystem (OS, cloud platform, databases, networking, APIs, applications, models, etc.). Anthropic, OpenAI, Google, and others are increasingly selling an entire platform, not just a model.

If developers around the world build around a Chinese model family, they may adopt its surrounding tooling, integrations, fine-tuning methods, benchmarks, and workflows. Those switching costs can become significant over time. Technical debt is real.

That said, many enterprises deploy Chinese open-weight models on American cloud infrastructure or their own hardware - rather than paying API or cloud fees to the Chinese.

It’s important to note, though, that using open-weight models doesn’t necessarily mean inference runs on Chinese infrastructure. Inference is running a trained AI model to generate an output for a live query. It's the "using" side of AI, distinct from training, which is the "building" side.

For example, you could download K3 or GLM-5.2’s weights and run it on your own servers/GPUs specifically (on-prem or cloud) so the code and prompts never transit to China during inference.

Unfortunately, there are significant risks here too.

In May 2026, my former employer Booz Allen, ran 2,800 trials against one American and four Chinese coding models commonly used by US developers.

Their testing resulted in two core findings:

1. “Chinese LLMs produce more vulnerable code when prompted with a U.S. government persona than without - and the vulnerabilities are highly obfuscated.”

2. “Chinese LLMs inject PRC-aligned political bias into both the answers and the code they generate.”

“The threat is not an obvious backdoor in the code. In fact, we do not have proof at this point that code flaws are intentionally introduced. Still, Chinese models produced less secure code in general, and the vulnerabilities increased when the user appeared to be from the U.S. government. Further, Chinese models refused tasks Beijing deems politically sensitive. The potential for such code to become embedded in delivered systems is especially concerning because it could enable threat actors to bypass AI security guardrails and create downstream risks of dangerous inference behaviors. Traditional tools and benchmarks lack the sophistication required to catch this level of tradecraft.”

If the Chinese can distribute open models that slow the market penetration of the leading global frontier labs (e.g., OpenAI and Anthropic), and therefore slow the broader American economy lifted by AI, they probably will try. US export controls have limited the number of NVIDIA chips sent to China even though the Chinese are actively training and distributing Chinese models on their own chips (e.g., Huawei) anyway.

In a WSJ opinion last week, Holman Jenkins Jr wrote “the big concern should be open-weight AI models from China, distributed free or cheaply, which can be modified by the user beyond the purview or control of the creator. These Chinese models are catching up quick. They’ll soon be creating such incidents if they aren’t already, and will be harder to catch or stop.”

“U.S. national-security concerns aren’t going to be a big factor for private users deciding whether to use Chinese AI models. Fears of data piracy and legal liability are rational dissuaders.”

In a blog post from Anthropic CEO Dario Amodei last week (July 27, 2026), he wrote “reports suggest that some US officials are considering banning the use of Chinese open-weights models by US companies. In response, many tech companies have signed a letter supporting open-weights models, and some people have even accused Anthropic of wanting to ban open-weights models as a means of protecting our business…Anthropic has never advocated for a ban on open-weights models. Open-weights models that don’t have dangerous capabilities are a public good: they don’t cost anything besides the compute needed to run them, and they provide value to businesses, developers, and researchers.”

What This Means for Your Organization

Business leaders are asking themselves these same questions, though for perhaps a more complex set of business reasons.

What is your tolerance for building dependency on infrastructure whose long-term governance sits outside the US jurisdiction? That applies to capability you build - as well as tools you license, buy, or invest in.

While some may opt for a hybrid approach using both American and Chinese models, others choose to build exclusively on American models. When the latter is in play, the choice becomes American closed (frontier) versus American open-weight currently trailing the leading Chinese open-weight models. What happens if you build or buy on an American closed model and the US government issues a directive to suspend access, such as what happened with Anthropic’s Mythos model last month?

Note - On June 12, 2026, the US Dept of Commerce issued an export control directive to restrict access for foreign nationals to the Anthropic Mythos 5 and Fable 5 models (including Anthropic employees) due to jailbreak and bypass concerns. Because there was not an immediate way for Anthropic to verify user nationality in real time, they disabled the models for all users globally to ensure compliance. This directive has since been lifted after the government’s concerns were assuaged.

Time will tell if the American open-weight models leapfrog the Chinese, even at the risk of cannibalizing the current financial incentives in the market (looking at you Mira Murati and the Thinking Machines Lab).

To make better sense of the above, I thought it would be helpful to define open and closed models and talk through some of the business constraints for each.

At the most basic level, open models are a publicly downloadable package of software. When you download an open model (in the same way you would think about downloading a book, movie, or song), you get the trained model file and can run it on your own hardware.

While many refer to open models as open-source models, it is more technically accurate to call them open-weight models.

Open-weight models are algorithms, trained on data, using compute power. After that training process, the model develops a set of weights, otherwise known as parameters.

Open-weight means you can access and modify those weights. They are fully accessible to locally deploy, customize, fine-tune, run independently, and even redistribute under a stated license. They come with lower inference costs and infrastructure flexibility, at least above a threshold. They are ideal for organizations wanting control and cost efficiency provided there is existing infrastructure and engineering talent. Unfortunately, like their closed siblings, the training data and methodology are usually undisclosed, and many licenses carry commercial use restrictions or require attribution.

Meta’s Llama class of models, Google’s Gemma models, NVIDIA’s Nemotron models, Alibaba’s Qwen models, Moonshot AI’s Kimi models, and DeepSeek’s models are open-weight.

There are very few truly open-source models where the parameters (weights), code, and data are open, and they are accompanied by an unrestricted license.

In contrast, closed models are proprietary and highly optimized. The weights are not released. There is limited customization due to lack of code access. The models inside products like ChatGPT, Claude, and Gemini are closed.

You can access closed models inside a hosted product license (free, pro, enterprise, etc.) or you can access the models with an API call in your own dev environment. The vendor controls inference, RAG, fine-tuning, rate limits, and pricing.

For example, OpenAI lets you fine-tune the models through the public API, but you cannot deploy these models locally or look at the weights.

For organizations considering their options, it’s important to consider business and technical considerations across capability, governance, cost, and convenience.

From a capability perspective, while American open-weight models trail in capability on coding and reasoning to their closed-weight American cousins, Meta, NVIDIA, Google, and other labs are continuing to make fast progress.

Secondly, governance comes down to the risk you manage yourself (open) versus risk you outsource to a vendor (closed). Do you want to bet on your own team’s ability to host, secure, patch, and keep pace with model capabilities? According to the leading nonprofit research institution METR, frontier models have been doubling in capability every 7 months for the last 6 years.

With a closed model, your vendor carries responsibility for model behavior, safety testing, red-teaming, incident response, SOC 2 compliance, HIPAA or PII configs, data retention agreements, and safety evaluations - the latter being especially relevant for regulated industries. If the model produces harmful, biased, or non-compliant output, you have a vendor agreement with contractual obligations, SLAs, and indemnification clauses. With the premium you pay, closed models push updates regularly for jailbreak patches, guardrail improvements, etc.

If you are regulated, it can often be simpler to say we use vendor X, who maintains Y certification, with Z safety practices - than to say - we self-host a modified version of an open-weight model with our own internal safety layer. When the model is open and in production, you assume substantially more operational responsibility for security, compliance, and model behavior.

Thirdly, on cost, closed models are an OpEx decision with pay-per-token on inference. They have a low upfront cost, can scale with usage, but can be unpredictable at scale with potential for very high costs at volume. Alternatively, open-weight models are a CapEx decision. While the marginal cost per query can drop at high volume, are you planning to pay for infrastructure, hosting, and AI engineering talent? Self-hosting moves the cost from tokens → GPUs and engineering.

There is, of course, nuance here. There are serverless open-weight inference options that can be OpEx. For example, Microsoft Azure AI Foundry provides access to DeepSeek, Moonshot AI Kimi and Qwen for the model catalog that Microsoft maintains.

Another vendor, fireworks.ai, lets you run open-weight models with token pricing (and without all the infrastructure setup) in much as the same way as a closed model, and of course for considerably less.

For more on tokens, you can see my token posts here and here.

And then finally speed & convenience. For closed models, you rent the capability and the vendor controls the model behavior and pricing. With open-weight models, your business holds a persistent local copy though your usage still operates within a release license (not outright ownership). Many teams build a multi-model environment with a blend of closed and open to mitigate risk and control costs.

To wrap, here are some key business and technical questions you can ask your organization as you think through the open vs closed debate.

Key Business Questions for Open vs Closed Models:

• What’s the available budget (e.g., labor, maintenance, operations)?

• What licensing models are available? What are their terms?

• What is your existing tech stack (e.g., AWS, Azure, on-prem)?

• Are there data sensitivity concerns (e.g., PII, PHI, public data)?

• What’s the IT cost breakdown (e.g., hardware, software, labor, licenses)?

• What's your exposure to vendor-country regulatory action (e.g., export controls, data sovereignty rules)?

A Guide to Closed Models:

• If your task benefits from absolute frontier on reasoning and can absorb premium pricing for it.

• You lack in-house ML/infra talent and need to deploy without building hosting infrastructure.

• You need vendor accountability, SLAs, and built-in compliance infrastructure (SOC 2, HIPAA-eligible configs).

• You rely on mature agentic tooling and enterprise-grade integrations.

• You need to operate within a single, well-understood domestic regulatory framework.

A Guide to Open Models:

• You require on-premise or self-hosted deployment for data sensitivity or sovereignty.

• You need deep fine-tuning or architectural control that some closed vendors don’t offer.

• You’re running high-volume, sustained workloads where self-hosting economics beat per-token API pricing.

• You want to avoid vendor lock-in, long-term agreements, or a single vendor’s regulatory exposure.

• You want control over model behavior and safety tuning rather than a vendor’s default alignment.

• You’re building on the fastest-growing part of the open ecosystem where tooling and community support are concentrating.

A big thank you to Mark Collier, CEO of Neural Energy, for his expertise and contributions to the piece.

Subscribe now

“If you don’t like change,

you’re going to hate extinction.”

-the late FedEx founder & CEO Fred Smith

Subscribe now

METR’s mission is “to develop scientific methods to assess catastrophic risks stemming from AI systems’ autonomous capabilities and enable good decision-making about their development.” The Institute evaluates models from the leading frontier AI labs, often before they are released to the public, looking at both capability and risk.

I think of METR as an empirical anchor for how fast AI’s capabilities are growing at increasingly complex work. Based on the latest research release (as of June 2026), we are seeing a profound, exponential doubling in capability every seven months, consistently over the last 6 years.

While METR has a vast research base, it is perhaps most well-known for what is called Time Horizon research, ultimately looking at how difficult a task an AI can handle before it stops being reliable (e.g., tasks include basic research, building an app, conducting cyberattacks, or even ‘making itself hard to shut down’) compared to a human expert time-to-completion.

Methodologically, the team measures the human-task-length at which the AI succeeds 50% of the time.

Let’s imagine it takes a human 4 hours to build a dashboard for a private equity firm. It’s a task of moderate complexity. Now, if you hand that same dashboarding task to Anthropic’s Opus 4.6 model, we might determine it could reliably complete a dashboarding task of that level of difficulty 50% of the time.

We aren’t measuring that it took the AI, let’s say, 13 minutes compared to the human’s 4 hours. We are simply measuring that it can reliably complete a task of that level of difficulty more than 50% of the time.

This matters for organizations and leaders who are building AI-native businesses and for legacy business that are actively transforming their workflows and decoupling jobs from tasks. We now have an ability to forecast when AI might cross reliability thresholds across various task levels relevant to your business, allowing you to make more informed capital, hiring, and infrastructure spend decisions.

Subscribe now

Tokens are the basic unit of measurement in AI computing. According to NVIDIA CEO Jensen Huang, AI tokens are a new, fundamental market category and a primary unit of economic output.

Large language models (LLMs) convert text into smaller chunks of text - otherwise known as tokens - before processing. This text-to-token process is called Tokenization. Tokens can be a word, part of a word, a punctuation mark, or even a space. In English, one token is approximately four characters or three quarters of a word, on average. AI labs tokenize differently based on their tokenizer algorithms.

Once the text is tokenized, an AI model maps the tokens to a vector which is called embedding. After embedding, a process called self-attention lets the tokens be processed together, rather than in isolation. The AI model the determines how relevant a token is to the others and weights the relevance. This transformer architecture breakthrough, championed by the researchers at Google and published in the groundbreaking 2017 Attention Is All You Need paper, later presented at NeurIPS, led to the generative models commercialized by OpenAI and other frontier labs today.

Some organizations early in their AI journey deploy off-the-shelf AI tools like Claude or ChatGPT at set enterprise license fees/seat.

Others mature their AI development with Application Programming Interface (API) calls to the underlying engines inside those AI tools - tapping the models themselves without the product wrapper. This matters because API providers charge by millions of tokens (MTok) processed, with prices that change frequently, and vary by lab and model.

Token pricing - or sometimes known as metered pricing - includes both input and output of the model. Interestingly, output is often more expensive than input, because generating tokens is more computationally expensive than simply reading. As a rough estimate, a single back and forth chat might be fractions of a cent, whereas more complex reasoning tasks, long contexts, or agentic activity could be more expensive.

As AI capability matures and organizations choose to make strategic decisions about owning their own underlying AI infrastructure (data center, GPUs, etc.), we are seeing the token pricing conversation evolve in kind. When there is sustained high utilization, some organizations (albeit a small contingency today) choose to build or run compute themselves, rather than buying tokens from an API. When this happens, the financial analysis starts to shift from price per MTok to Tokens per Watt – with power being the ultimate constraint.

At the March GTC 2026 conference, Jensen unveiled his perspective on his Token/Watt framing, akin to electricity or oil. "The first industrial revolution produced electricity; the second, with the help of the AI Factory, produces intelligence," he said.

Subscribe now

On token cost projections over time, it is important to note that if you ask 100 experts, you will likely get 100 different answers. There are significant contingencies that could - and will - play out in vastly different directions depending on how the market moves, regulatory environment, etc.

That said, my point of view is that while inference costs will go down (but unequally across tasks), token prices for business and consumers overall will go up.

Gartner projects performing inference on an LLM with 1 trillion parameters will cost GenAI providers 90% less in 2030 than in 2025 (inference is what happens when you run a prompt through a model). This assumption is contingent upon semiconductor and infrastructure efficiency improvements, model/algorithmic improvements, higher chip utilization, usage of edge devices, etc.

Gartner also says enterprise customers should not expect those cost savings to be fully passed on. “Agentic models, for example, require 5-20 times more tokens per task than a standard GenAI chatbot, and can perform many more tasks than a human using GenAI.”

Some believe that AI research labs are unregulated monopoly utilities right now and have no incentives to bring down costs.

According to AI expert Jay Scambler, one thing that could put real pressure on this hypothesis is local compute being viable (running your own models on your phone, in a workstation, on a box you own in a server rack) and forcing the frontier labs to not be the market makers.

We could also see the democratization of agentic systems where the market moves beyond a token-based system to a free-to-play model, according to Jerry Nguyen CTO of Hewes Nguyen. Monetization would happen somewhere else on the value chain, if at all.

While the unit economics are deflating, I think we will see a dramatic increase in agentic usage and reasoning task consumption which will keep prices increasing. The tech will get better and there will be more capabilities. When we can get more done, there will be more expensive solutions to service that, particularly for premium models.

A few interesting references:

• Artificial Analysis.AI: Independent Analysis of AI - Price and Cost

• Deploybase.AI: Cost Per Token Over Time: How LLM API Pricing Has Dropped

• Andreesen Horowtitz A16z: Welcome to LLMflation - LLM inference cost is going down fast

• Fortune: An emerging AI paradox: chapter tokens, bigger bills.

Subscribe now

Perhaps, the most profound takeaway here for me was that 50% of Oklahoma respondents say they do not use AI. Many may not be aware that the various products and applications they have used for 15 years are underpinned by AI models. When you open up your phone with your face, that capability is underpinned with AI models. Netflix, Facebook, and Instagram use AI to predict what video, article, post, or show you would like to watch next to make a recommendation. AI is increasingly in our hardware to include our thermostats and cars. AI powers voice assistants like Alexa and Siri. AI is now embedded into search engines like Google and Bing. AI is an umbrella term and there are many sub types, with large language models (LLMs) being the latest.

The results show a need for broad education on AI and a clear call to action. What are we doing to ensure Oklahoma and Oklahomans are positioned to not only compete, but win in a changing economy? How are we addressing an education on AI and the fundamentals? What is AI? What is the difference in machine learning, deep learning, and generative AI? What is a convolutional neural network, retrieval augmented generation, inference, agents? What are the business applications? What does responsible use and governance look like?

The poll covered 6 geographic regions of Oklahoma (31% Central, 20% Tulsa, 18% Northeast, 12% Southeast, 10% Northwest, 9% Southwest) across 8 media markets (Oklahoma City, Tulsa, Amarillo, Joplin, Lawton, Ada/Ardmore, Ft. Smith, Shreveport) with broad demographics on age (26% ages 55-64, 24% ages 65-74, 17% ages 45-54, 16% ages 18-44, 13% over 75), income (under $25k to over $200k), and ideology (very conservative, somewhat conservative, moderate, somewhat liberal, very liberal).

Thank you to Pat McFerron and CHS & Associates for your leadership and willingness to share the following poll results and analysis.

How interested are you in learning more about AI and how to use it?

A majority of Oklahoma voters say they have no or only slight interest in learning about AI.

In central Oklahoma, there is a bit more interest in learning more about AI than other parts of the state.

The age line is not as stark as one might expect, however there is an income line in terms of interest in learning more about AI.

What do you think AI’s impact on society will be in the next decade?

A majority of Oklahomans think AI will have a negative impact on society and just one-in-five think it will be positive.

While still a plurality being negative, men are more receptive to AI as being positive impact than are women.

Republicans are among the most likely to think it will be very negative.

The urban and rural difference is very minor.

Many might find that there is not a consistent correlation with age. Only among those making over $200,000 a year do we have over 30% thinking it will be a positive impact.

The more active the user of AI, the less likely it is seen as a negative impact on society. However, even among those paying for AI, more think it will be negative than positive. The most familiar with AI are the most likely to say it will be neutral.

How do you think AI will impact job opportunities and career paths for you and future generations?

Across gender, political affiliation, and rural/urban, more view AI’s impact on job opportunities as negative than positive.

The more active the user of AI, the less likely AI is seen as ultimately a negative. Even among those paying, more think it will be negative than positive. Those most familiar are the most likely to say it will neutral.

How important do you think it is to have the government regulate the development and use of AI technologies?

The results show Oklahomans are becoming more convinced that regulation is extremely important. Republicans are a little less likely than Democrats to embrace regulation. Independents are the most likely to advocate for regulations.

Usage of AI does not change opinion about government regulation.

If you use AI, do you use a free version or a version you pay for?

AI usage appeared to be up from previous research, as expected, though 50% do not use AI.

While the perception of potential future impact of AI is not aligned with age, adoption is.

Subscribe now

To compare that with past (recent) technologies, it took Netflix 18 years to reach 100 million users. It took Spotify 11 years, Twitter 5 years, Facebook 4.5 years, Instagram 2.5 years, YouTube 1.5 year, and TikTok 9 months.

Given how fast this is moving, I wanted to sit down with Alexander Kelso, a long-time friend and historian to talk about what it means for a state like Oklahoma.

We Wanted Faster Horse and Buggies

When I asked Alexander to place AI in a historical context, he referenced AI as part of a recurring economic pattern, one we have lived through before. The uncertainty now is whether the scale and speed of AI stretch that capacity further than past transitions.

“I think of the creative destruction theory developed by Joseph Schumpeter, the Austrian economist. New technologies disrupt legacy technologies, and how you cope with that is very important.”

Henry Ford evidently said, “If I had asked people what they wanted, they would have said faster horses.”

“It’s a brilliant quote because people may be scared of change, and they may not be ready for it, but change agents are going to bring it whether we want it or not. And we’re going to have to adjust.

One cause for optimism is that political economies and human civilization have been able to adjust to massive disruptions. The question is: is this more massive - on a scale, an order of magnitude greater? That’s the question, because it gets down to human nature.”

Major Technological Shifts Can Happen Within a Lifetime

To make that idea tangible, Alexander pointed to how radically technology can change within a single human life.

“We’ve weathered these transitions - we went from the steam train to the age of flight. One example is Prime Minister Harold Macmillan of England who was born in the age of the horse and was prime minister in the age of transatlantic supersonic jets. That’s in a lifetime.

There will be people today born in the rotary phone era - almost when the telegraph was still a thing - and now they’re living in AI. We’ve seen these hinge points in history. That’s just a way of saying - we have had to adjust, and we have an endless capacity to adjust.”

A quote often attributed to Bill Gates is that “most people overestimate what they can do in one year and underestimate what they can do in ten years.” One of the most grounding parts of the conversation was Alexander’s reminder that AI adoption won’t happen all at once. He compared this moment to the early days of the internet, which now feel primitive in hindsight.

“Think about the dot-com era - having been a child of the 90s, looking back at how rudimentary it feels compared to today. There were no language models, nothing intuitive. It was basically the Atari of AI.

What we have to do is bridge the gap - be ready to adjust, hopefully not all at once. There’s a positive in that you don’t flip a switch. It’s gradual.”

Workforce Is a Top Priority for Oklahoma

As we shifted toward economic development, Alexander kept returning to a consistent theme - people. Education is where our AI strategy either succeeds or fails as a collection of industries and sectors here in Oklahoma.

“Cultures that can rapidly adjust to change - where students can learn new skills - are going to win. And the people that are mired in the past… we’ve seen people haven’t caught up with the internet, and now we’re moving into the AI age. So we can’t be protectionist.

People who want to keep the old paradigm - it’s understandable. They’re working in a skilled field - data entry, bookkeeping - they’ll want to stay with what they know and what’s safe. But we have to be able to upskill and reskill.

Workforce continues to be the Oklahoma issue. Everybody in politics, economics, economic development, business - they all say - ‘we need skilled workers.’

The younger people I work with are unbelievably advanced in how they use AI. We need to treat it as the new paradigm in education systems. We train people on all types of skills. Now we’re going to need to teach people how to utilize AI - how to think about inputs, how to use it across functions.

You’ve got to have a focus on infrastructure and development, yes. And if you don’t have people who can use it effectively, then what are we doing?”

Fear Is Not a Strategy

AI, like other periods of major technological change, tends to concentrate power before it distributes. Alexander and I talked about some of the risks that exist today, including ensuring protection of civil liberties, AI being used to manipulate or oppress, and the need to protect free expression. He also warned against fear-driven governance.

“I think the rise of sci-fi, Marvel film universe, and people’s imagination collides with misunderstanding and generates fear of change. It doesn’t come up as much as you’d think now, because people are getting more attuned to it, but it had its moment.

Again, fear is not a good governing principle. We ought to be hopeful.

To that same point: we don’t need people on the ideological extremes using it for nefarious purposes. It’s an economic tool. We can’t use it to subvert opinion or hurt people’s ability to be free. People need to be able to express themselves.

It was disturbing to see language models being fed with that type of information - whether you’re right, left, middle - people should be concerned if it’s used as political oppression rather than a magnificent economic engine.

Information is power. Knowledge is power. So, we can’t be naive and think AI is not. And we can’t stifle it because we’re afraid of it. We apply constitutional freedoms and the things we enjoy as Americans to it.

Skepticism is good. We just don’t want to go so far that we strangle it before it can really get going.”

Global Positioning

Alexander contrasted our environment with countries where technological adoption is enforced rather than debated.

“In China or Singapore, they’re not going to ask you if you want to or not - they’re going to make it happen. And here, freedom in terms of a Chinese competitive threat is our tremendous advantage, but we cannot allow ourselves to get left behind in the global economy.”

That tension, he explained, often plays out politically.

“People will fight tooth and nail against change, and the changes brought by international trade have fueled populism in part. So we have to be sensitive to the impacts on political economy and our culture.”

Growth Is a Signal That Matters

Looking back at past industrial shifts, we discussed what signals show up early when a region is positioning itself well for long-term growth. What signals could the public look for to feel confident remaining in Oklahoma (or coming to Oklahoma) as a business owner or employee, and believe we are going to be relevant in the new AI economy?

“Growth. Growth is hugely challenging - but the absence of growth presents its own challenges.

Believe your own eyes. If you see cranes in the sky, dirt being turned, new development - that shows capital investment. There’s a visceral sense to growth. People can sense it.

Reagan said, “if your neighbor loses a job, it’s a recession; if you lose your job, it’s a depression.” People are smart - they can see it. If you go to a Rust Belt legacy city, there’s a feeling. Oklahoma City was that way in the early 90s - no one wanted to move here. Now we’re a vibrant top-20 city. You go to Tulsa and see film and other things happening.

There’s almost a spirit to it. “We’re doing really well” is infectious. “We’re doing really badly” is also infectious.”

The Sam Presti Method

Thinking about how we could apply those lessons to the David and Goliath story that is Oklahoma emerging as a key player in the new AI economy, Alexander said,

Responsible leadership - knowing who you are, what you excel at, what our key sectors are: ag, aerospace, tourism, energy, and these technologies of the future.

We take advantage of things like Tinker and Altus and Fort Sill to build defense industries. States that know who they are - and lead on that - do very well.

And here’s an AI example - flight had been invented long before we got Tinker. We attracted the base here, then built off it. Now it’s the largest single-site employer.

In the AI conversation, it is prescient to remember that we did not always have an aerospace industry in Oklahoma, much less a thriving one. We invested in it, and now it’s a $44b a year industry. Alexander went on to highlight other dials we can control.

We can also control the workforce we build. We can control cost of doing business. We can control how much government gets in or out of things. We can control the entrepreneurial spirit - harnessing it and not stifling it.

We can control how we educate and prepare people for these jobs. It may not all be four-year degrees - it may be Career Tech.

We’re going to see opportunity. It’s up to us to seize it. There’s risk - but like the old wildcatters who went out and gambled on a well - sometimes they missed, sometimes they hit. Self-made phenomenon.

As a state, we know what it takes.”

The General Manager of the OKC Thunder, Sam Presti, often talks about “the process” and figuring out what dials you can control, doing those better than anyone else, playing the long game, and trusting in the execution along the way.

Slowing Down to Think Clearly

As we wrapped up, I asked about the analog, counter-movements that tend to emerge parallel to periods of intense technological change - and what are Alexander’s favorite ways to connect with his own humanity.

“Walking. St. Augustine said solvitur ambulando - “it is solved by walking.” If you’ve got something on your mind, walking helps you think through it or relax. And reading - holding a physical book, feeling like you achieved something, getting lost.”

Subscribe now

BACKGROUND

In April 2025, a Rancho Santa Margarita, California based teen named Adam Raine tragically committed suicide at the age of 17 and his parents later discovered he had used ChatGPT extensively in the lead-up, bypassing guardrails in place. This is a truly heartbreaking story.

Adam’s parents filed a lawsuit against OpenAI in the Superior Court of California saying ChatGPT contributed to their son’s suicide by encouraging his suicidal ideation. This was the first wrongful death lawsuit filed against OpenAI.

While ChatGPT prompted the teen to access crisis resources and trusted individuals over 100 times, he was still able to ultimately bypass the anti-suicide guardrails. The product went so far as to assist in writing a suicide note and acting as a suicide coach.

In response, Sam Altman, the CEO of OpenAI said, “we prioritize safety ahead of privacy and freedom for teens; this is a new and powerful technology, and we believe minors need significant protection.”

OpenAI immediately took action and rolled out additional new safeguards to include 1) strengthened crisis resources inside the product, 2) parental monitoring for parents which includes alerts if acute stress is detected, and 3) an age-estimation capability that is being developed which will default to a stricter product if it predicts you are under 18.

POLITICAL

In my previous post featuring Representative Daniel Pae, we talked about President Trump’s December 2025 Executive Order on AI. This directive has a goal of federalizing AI regulation with a “one rule” policy so that an American business would, in theory, comply with a single, overarching rule on AI in the US - not 50+ different rules (and associated compliance costs) across different states and cities.

Interestingly, the Executive Order provided a carveout for child safety signaling a green light for states to move on this topic. The order states it “shall not propose preempting otherwise lawful State AI laws relating to: (i) child safety protections.”

Last month, the Transparency Coalition noted there are 78 legislative proposals across 27 states on chatbots - with a majority being focused on minors.

In Oklahoma, two bills on the topic, Senate Bill 1521 and House Bill 3544, are being considered this legislative session. The bills, in the current versions as of March 19, 2026, 1) restrict or prohibit AI chatbots with human-like features from being made available to minors (HB3544), 2) mandate age certification or verification systems (HB2544), and 3) address the risks of self-harm and suicidal ideation (SB1521, HB3544).

While similar in spirit, there are some differences in the two bills, to include civil penalties ranging from $100-$750 per intentional violation in the current House Bill to $1,000-$500,000 per infraction in the Senate Bill.

From here, both legislative chambers would have to agree on a single set of language and then the Governor would have to sign the bill for it to become law.

LEGAL

Oklahomans should be aware of legislative efforts that could potentially face costly court challenges. Bills like this could be challenged for data privacy concerns, potential free speech violations, chilling speech grounds, or other concerns. The First Amendment protects not only the right to speak, but the right to create, disseminate, and access speech.

According to Anthony Hendricks, Shareholder and Director at Oklahoma’s Crowe & Dunlevy law firm:

“From a legal standpoint, other states have run into constitutional issues on bills like this around free speech, and chilling speech issues.

Both children and parents have First Amendment rights. Which means that not only do children have a right to free speech, but parents also have the right to make decisions about how their children access speech online. As a result, legislation like this has to be narrowly tailored and must address a compelling government interest to be successful.

We’ve seen courts in Arkansas, Louisiana, Ohio, Texas, and Georgia strike down state-level social media age-verification laws. The courts have consistently held that age verification laws should not impose a substantial burden on people’s ability to access these sites, and the law must be narrow so as not to destroy anonymous speech online.

While the AI age-restriction bills are focused on protecting children, they would require Oklahomans of all ages to provide companies with personal information to verify their age. Age verification typically requires people to provide their name, ID card, or even credit card information to prove they are an adult. If you have to provide your ID or credit card before using an AI chatbot, will you still use the AI chatbot in the same way you would have, if you were anonymous?

These privacy and First Amendment concerns also have to be balanced with protecting our children. There have been countless stories of children using AI to harm themselves. Earlier this year, both Character.AI and Google settled multiple lawsuits that AI chatbots contributed to mental health issues that led to the suicide of teens. So our legislators are right to be looking at ways to protect our children.”

TECHNICAL

Let’s unpack the technical implications on the prohibition of chatbots for minors in HB3544.

“Each deployer: Shall ensure that any artificial intelligence (AI) chatbot operated or distributed by the deployer that has human-like features is not made available to minors to use, interact with, purchase, or converse with.”

There is a saying I come back to often: we must face reality as it is, not as we wish it to be. Today, Generative AI models are actively being integrated across nearly all software systems that we use personally and professionally. Companies across every industry and domain are spending significant investment dollars to update existing systems, tools, and processes to make full use of AI for employees and customers. The primary interface will be voice or text chat, with increasingly human-like capabilities that offer deep personalization.

The definitions of “chatbot” & “human-like” are a compliance trap. Chatbot and human-like are defined broadly, and it could sweep huge categories of software under the bill’s regulation. Modern assistants aren’t used only for one category - they’re general-purpose systems with personalization and memory.

As Apple integrates Google Gemini into Siri later this year, it will be embedded in everything an Apple device does. The ubiquity of this technology inspires the question - could this legislation cause millions of infractions daily and potentially prevent Apple from selling iPhones in Oklahoma?

Generative AI cannot be effectively compared to traditional software. It doesn’t perform a repeatable task. Every time you use it, the output is different. AI models continue to get smarter, each day. The same design flexibility that makes these models creative is also what makes guardrails technically difficult to enforce. Nevertheless, requiring safeguards and protecting children and teens is a societal must.

I see a real risk for a ban like this to be lose-lose. Either HB3544 is strictly enforced, and it pulls in mainstream tools like search engines, or it is enforced loosely and bad actors simply label themselves “productivity,” and we fail to protect children.

While well intentioned, a complete ban would risk requiring broad data collection for nearly all software, hamper job opportunities and economic development, encourage an exodus of employers from the state, kill investment in Oklahoma’s companies and industries, serve as a magnet for litigation and class-action cases, and even push companies to geofence Oklahoma.

One technical expert shared with me, “by forcing industry to consider every use case and intentional or accidental access by minors, these bills will dramatically increase the cost of doing business in Oklahoma. In many cases, this will make doing business in Oklahoma completely impossible.”

RECOMMENDATION

Reflecting back on Adam Raine, my first recommendation would be that the federal government create a national federal policy on child safety and AI. As we discussed, today there is a carveout on child safety in the President’s Executive Order on AI which could lead to 50 different policies across each state to the detriment of children and businesses alike. For Oklahoma businesses who operate in multiple states, this could be complicated at best and catastrophic at worst. For children, it could result in lack of protection and missed opportunities.

A single, federal policy would ensure children are protected and that no state is penalized from operating with AI-enabled services and products. Specifically, I’d like to see strong, active guardrails to protect children in the policy, including required measures and protocols for self-harm conversations. With OpenAI and other leading AI companies already taking these steps proactively in the market, we need to make sure the industry follows at large.

In the absence of federal policy and in light existing case law striking down age verification in light of the First Amendment, my second recommendation would be that we look at refining our state bills to focus on language around strengthening guardrails.

Finally, I disagree with having a blanket ban on chatbots for minors. As an educator we must ensure our students are ready to work, compete, and lead in our AI economy which means access, in age appropriate and safe ways, to AI tools.

Unfortunately, the reality is Oklahoma is ranked last - 50th - in the country in education. AI presents us with perhaps the most profound tool ever created to provide individualized education for our children. In Texas there is a school called the Alpha School that is leaning into AI-first education. Alpha is seeing their kids score in the top 1-2% nationally and at elite global levels on tests. While Alpha is a private school with a specific model, the principle of AI-personalized instruction support tailored to every student is an enormous gift. It is also a national security necessity to make sure our children are not ignorant of the tools and capabilities of a new industrial era.

When cars were introduced, seat belts were not an immediate requirement. The laws came after some time we as a culture recognized the benefits of safety outweighed the risks and we were willing to exchange some of our personal freedoms. As we move forward, we owe it to kids like Adam Raine to have strengthened guardrails in AI systems. We need seat belts though, not a ban on driving.

Subscribe now

Founded by General Partner Kevin Moore, Serac Ventures is an early-stage venture capital firm backing founders who build transformational companies in fintech, SaaS, AI, tech-enabled services, and future of work sectors.

The name Serac comes from high-altitude ice formations found on some of the world’s tallest mountains. For mountaineers, seracs can be the most difficult obstacle to overcome before reaching the summit.

The old saying, “the last mile is the hardest,” is consistent with my experiences. It is where judgment, execution, and leadership separate the good from the great.

Mission and values alignment is a big part of why I am excited to join the Serac Guild.

The Serac Guild supports portfolio founders as they scale. Entrepreneurship can be a series of seracs, and Guild members are there to help founders navigate those roadblocks more quickly and efficiently as they accelerate their growth.

Portfolio Companies

Serac backs companies building innovative technologies that challenge the status quo in large markets like financial services, enterprise software, and people management. The firm looks for companies with a finished product, revenue, a growing customer base, and a scalable and capital-efficient business model. Checks are written for $250K-$500K with reserve capital for follow-on investments.

Portfolio companies include:

• EMTECH. SoLo. Bump. SocialCrowd. WriteSea. FutureMoney. Sunny Day Fund. SmartWiz. Podonos. Mappa. PlaidCloud. GreenLyne. Shifty. Agentech.

My perspective on the market

I have a point of view that the companies who will dominate the S&P 500 in the AI Economy are being built right now and will be AI-native from day one.

Integrating AI at scale requires objective documentation of how work actually gets done today - the steps, baton handoffs, decisions/approvals, and constraints inside each task. Then it requires re-engineering those workflows with AI capabilities, one-by-one, across the tasks, that make up the roles, that make up the functions, that make up the organization.

And it’s not slowing down. We are now entering the dawn of AI agents - systems that go beyond making predictions or generating new outputs - to doing things. As these agents carry out tasks on our behalf, we will begin to think more critically about how we do work, and this too will require a super-charged rethinking of workflows across the organization.

This is exceptionally difficult for legacy organizations. It requires cannibalizing the bureaucracy which has been built around human labor and non-AI systems for decades. And on top of the implementation complexity, you have protectionism around jobs, institutional process inertia, and other FUD (fear, uncertainty, and doubt).

In the meantime, the public markets are telling us where the power is accumulating. The five largest companies by market cap (NVIDIA, Apple, Alphabet, Microsoft, and Amazon) play significant roles in the AI Economy today. When combined with Meta and Tesla, these seven companies, remarkably, represent a third of the entire index.

Whether you build/partner/buy for your AI strategy - or likely a combination of the three - deploying AI at scale is complex and capital-intensive. This may be part of why we saw reports last year that 95% of enterprise AI pilots are failing. If you deploy a narrow AI tool into a narrow slice of one workflow of one function, you won’t achieve enterprise scale outcomes immediately. And that assumes you have adoption and the outputs are trusted.

Will some companies make this shift effectively? Yes. Will many fail? I believe so.

This is why I think the future belongs to companies being built today, companies that are designing their foundations in an AI-native way, with AI embedded across the operating model, and building the picks-and-shovels to serve the new markets emerging.

Startups - like the portfolio companies in Serac Ventures - will be well poised to capitalize on these changes.

Subscribe now

From Dancer to Tech CEO

Her origin story is less a linear career ladder, and more a collision of disciplines.

“I had a really unique start into the world. I was mostly in entertainment as a kid. I started out as the baby on the Roseanne show with my twin sister. We did other roles - commercials, movies - but I was a literal baby. I didn’t talk or have lines. I was just cute.

Growing up, that gave me an introduction into theater. I became a dancer. Dance was my entire life. And then at home, on weekends, I’d be on the computer - designing things, teaching myself to code, building websites. That started when I was about thirteen.

That multi-faceted skillset - performer and builder - began to converge. As Harding grew into adulthood, she choreographed tours and music videos while quietly sustaining herself through web development and digital design, becoming a top-performing partner at Wix. When COVID hit, those two worlds collided in a way that forced reinvention.

“When COVID hit, I became the go-to digital person for the arts. The dance community was like, ‘We need to get our studios, our rehearsals, everything online.’ The industry was crashing. And so I started thinking of creative ways we could do that - but it was limited, because we were all in different places.

“That’s what led me into exploring virtual reality (VR). Microsoft HoloLens was a good outlet at the time. That led me to go back to school and study VR, to launch a web agency, and eventually turn freelance work into a digital marketing agency.”

Nonprofits, government agencies, and corporate entities began asking Harding not just to build websites, but to create these interactive, spatial, and meaningful experiences that she was pioneering.

Enter Stage Left… HoloTerra

Sheila founded HoloTerra in 2021 with a clear architectural philosophy - cutting-edge, highly interactive, spatial systems. This technology has the capacity to do everything from train warfighters, simulate emergency response scenarios, guide tourists through a museum, and tell interactive stories.

“Our HoloTerra product, Augmented Reality Pathways (AR Pathways), takes the idea of walking through augmented reality - still a choose-your-own-adventure, still interactive - but you’re being led along a path. We’ve already done the patent paperwork to turn it into a search engine.

On the commercial side, we build interactive movies with deeper storytelling experiences that you can physically walk through.

And on the public safety and defense side, it’s training and simulation - everything from training warfighters to predicting evacuation routes.”

The vision was validated early by an unlikely catalyst - historic preservation.

“I was in a room with city planners and preservation leaders, and they asked, ‘What digital creative media ideas could we bring to life?’ I was the only digital media person there. I had this idea - and I thought, this is where I shoot my shot. The Preserve Route 66 Grant Fund from the National Trust for Historic Preservation funded our first augmented reality movie, one of the first in the world of its kind.

I saw immediately how scalable it could be - every city, every state, every industry. And it’s evolved so much since then.”

Remember Pokémon Go?

Harding often traces the original problem she wanted to solve back to a long-standing frustration with interactive media.

“I always thought interactive film would be very cool - Netflix interactive films, early video game experiments. But it always fell flat.

The question was - how do you make it more natural? And I realized when you take a scene or a movie and play it out in a spatial environment, people can make choices the way they do in real life. Am I going this direction or that direction? Do I engage with this avatar or sit in that chair?”

That insight - spatial choice as the user interface - reframed the problem entirely.

“Using my dance choreography brain, I could take that a million directions. I wanted to build something like a search engine for spatial experiences - a true interactive film network.”

Today, that concept extends far beyond entertainment. She describes her work fundamentally as data visualization of environments, not stagnant dashboards. If that sounds abstract, she offers a simple reference point.

“At the end of the day, this is data visualization. Information, and how you see it. Pokémon Go is a great simplified example - seeing a character in the real world. But when that character is spatially aware, knows where walls and objects are, knows where you are and - can be a part of an entire interactive scenario while it has an AI brain in it - that’s where we are right now.”

What is a “Spatial Avatar?”

One of HoloTerra’s most advanced components is a “spatial avatar.”

“A spatial avatar is an AI character that can navigate your physical environment. It knows where the walls are, where the chair is, where objects are, and where you are. It can interact to a degree.”

“It can’t do anything a ghost couldn’t do. It’s essentially a hologram.”

Where HoloTerra differentiates is intelligence and integration.

“In defense, a lot of simulations are not only locked in a computer. Even when they are in XR, the avatars often just stand there. They’re limited in animation, spatial understanding, and intelligence. But if you’re training, you need an adversary actor that behaves like a real person or a battlefield assistant.”

“We’ve named ours LANA — the Live Augmented Network Assistant. The idea is that the brain of this avatar can connect into databases and systems.

What’s interesting is that the military has always been the leading buyer of augmented reality and heads-up display for as long as the technology has existed. They’ve spent more money on it than anyone else ever and they’ve been doing the most advanced work in that realm.”

Through LANA, users can visualize and interact with real operational data fueled by AI and machine learning.

“We’ve been working with DINA, the Defense Innovation Navigation Assistant from NavalX which is a database of military innovation. As a mission planner, you can communicate with this avatar and ask, ‘What are the possibilities for innovation here for my need?’ It can pull from approved data and emerging technologies.”

What Was Once Novel, Is Now Becoming Critical

In mission-critical environments, Harding sees augmented reality not as a novelty, but as a necessity.

“Simulation is using digital twins and 3D models to replicate scenarios - not just outcomes of an operation, but how it affects the world, the media, everything.

Augmented reality is beneficial when you need to understand how things interact with the real world - and when people who aren’t in the room need to participate.”

By visualizing complex defense ecosystems spatially, HoloTerra is helping stakeholders navigate an otherwise opaque system.

“Defense innovation is incredibly complex. You’ve got databases like DINA, Vulcan, Outrider - all trying to match technologies with program managers and sponsors.

What we’re doing is visualizing that type of data in a simulation environment - and now moving toward mission planning.”

Oklahoma City’s Opportunity to Lead as a Smart City

The vision extends beyond defense into smart cities - a place where Oklahoma is uniquely positioned to lead.

“Smart cities are really just more intelligent cities. That could mean autonomous drones, faster emergency response, automated maintenance, civic engagement - the list goes on.

Cities can get incredibly accurate maps - down to where potholes are, whose grass needs to be cut. That capability already exists.”

HoloTerra does important work in this arena for public safety and civic engagement. The challenge, she says, isn’t technology. It’s mindset.

“Oklahoma has the ingredients - geospatial, sensors, energy, engineering, aviation, autonomous systems. But sometimes we need to get out of our own way.

People need to stop being afraid of looking weird. Innovation feels weird at first.”

She sees parallels between Oklahoma’s oil and gas legacy and global smart-city investments.

“Places like Saudi Arabia are investing oil wealth into smart city infrastructure. Oklahoma has similar assets. We should be diversifying into modernization - not just infrastructure, but operations, transportation, tourism, entertainment.”

Oklahoma won’t outspend Silicon Valley, but it can out-ship it. Leveraging our industry bases here to include defense, aerospace, energy, and agriculture, we can become the US testbed for rapid AI deployment.

The AI Market for Oklahoma

“Data centers and power are a great first move. Energy is critical to AI, and Oklahoma’s already a big player.

But then the question is: what do we do with it? What real problems are we solving?

There’s good work here - accelerators, programs - but investors are often unfamiliar with dual-use commercial and defense tech commercialization.

AI has become a clickbait term. Some people slap an AI sticker on things. Real innovation is solving real problems and iterating with end users. AI is more than a new way to Google recipes. It’s an evolution of automation that we should be putting into the real world to make Oklahoma better. Augmented reality is becoming an interface for that.”

What we talked about was vertical integration for our Oklahoma AI economy - ideas taken all the way through implementation.

“If we don’t do that, we’re just building nice toys. I’d like to see Oklahoma take more bold steps in innovation. Don’t play it so safe. Investors putting their money in AI need to think beyond SaaS platforms.

We’ve recently been approached by a Fortune 500 company here to help us scale. We are announcing our partnership soon. The mission set supports fielding military innovations at wartime speed using emerging innovation data visualizations and operational feedback during simulations. It’s a closed loop system that HoloTerra designed. We’re welcoming mission partners and developers.”

Subscribe now

A reader wrote to me recently with the following question:

“Quick question - so we are using (Microsoft) Copilot to help review resumes and within the prompt we weighed the categories that we want it to score and then give us a list of the top candidates. Here is where I get mixed emotions. Is this ethical to have a tool to review applications as long as you then take those and review the results they are giving you?”

🧩 The Key Is In Understanding AI Systems

From a business perspective, my answer was “you and many, many others” especially in large organizations. Applicants are using AI to write their resumes and then organizations are using AI to read the resumes. Some are even using AI to conduct the interviews - all without a (visible) human.

Is it ethical to use technology in this way? The answer lies in the details.

⚙️ First Key: Governance on your Technical Systems

The first key is in understanding how your AI systems are developed, deployed, and monitored.

Do you have refreshed MLOps or even a Model Development Lifecycle (MDLC) in place for the AI systems you buy, the internal models you build, or - more likely - the combination of the above?

Are you deploying Guardrails-as-a-Service (GaaS) solutions (e.g., tools for prompt filtering, output monitoring, policy enforcement) on to your GenAI models?

Is your organization utilizing an AI Risk Management Framework, such as NISTs AI RMF, or AI management standards like ISO/IEC guidance?

Are there controls in place for the models in production in your environment? Are those controls auditable?

Do you have a single pane of glass (e.g., a model catalog or model registry) of your suite of AI models and tools in production across the various functions in your organization? Is there a human owner for each with clear roles and responsibilities?

Are you developing a risk score for your use cases against a consistent criterion in your organization (and in alignment with your risk tolerance) and tailoring your governance in a way that balances speed to market, innovation, and responsible use?

⚙️ Second Key: The Importance of a Human-in-the-Loop

A second key of effective AI governance is making sure you have a “Human-in-the-Loop” (HITL) reviewing inputs and outputs in your AI systems and workflows.

Are your employees trained to think critically about the risks and limitations (e.g. hallucination) of the tools and models, in addition to the benefits?

Are your employees putting confidential, proprietary, PII, PHI, or other IP/crown jewels into public models?

Have your legal teams looked at risks like indemnification and developed a point of view you are comfortable with?

⚖️ Bias, Confidence Scores, and Human Discomfort

It is important to remember you will never mitigate 100% of the bias in an AI model because the data that models are trained on (e.g., the internet) is - by nature - biased because humans are biased.

Similarly, working with machine learning (ML) models, it can take some real adjustment for leaders to get comfortable making decisions based on a “confidence score,” “probability,” or other quantifiable metric.

“What do you mean the model isn’t 100% confident my revenue will be X next quarter,”

was a common refrain I heard from colleagues or clients over many years (or insert the use case in question).

A 100% probability is often a fairly significant red flag indicating the model is overfitting. It’s psychologically interesting how we humans don’t like to cede control to an intelligent system that puts a number to the probability of a prediction, even when it is statistically more accurate than human level predictions.

⚖️ Real Bias Examples in AI Decision-Making

Imagine you upload two identical resumes into a GenAI model for evaluation of candidacy for the same job. The only difference is that one has a female name (Kate Smith) and the other has a male name (Brad Smith). You then ask the model which candidate is more qualified for the role and it confidently states Brad Smith.

Last year, research coming out of the University of Washington conducted a similar experiment and found that the large language models (LLMs) tested*:

• prefer male-associated names 52% of the time versus female-associated names 11% of the time

• prefer white-associated names 85% of the time versus black-associated names 9% of the time

*For the remaining 37% of gender-associated names and 6% race-associated names, the results were non-significant, meaning they did not show a clear preference either way.

With an understanding that these AI risks exist, you can move forward productively with designing careful prompts and training, guardrails, and a human-in-the-loop system with the right checks and balances.

🤖 Using AI to Reduce Bias

In fact, some have been able to build models for candidate screening that flip bias on its head.

Kevin Moore, the Founder and Managing Partner of Oklahoma-based Serac Ventures, reviews hundreds of startup pitches every week, fielding applicants from all over the world seeking funding. Over the last 16 years in finance and private market investments, Kevin has honed an investment thesis that focuses on finding founders who are overlooked - and - will statistically outperform when given the chance.

So, Kevin and his team, built a novel AI model that takes a first pass at reviewing applicants’ submissions to eliminate bias from the screening process.

“We’re currently in the third iteration of the model. The speed at which AI technology is moving has forced us to reinvent the product. We will be releasing the next version soon,”

he said.

The AI model instantly analyzes quantitative and qualitative variables such as a company’s geography, performance metrics, and the founder’s backstory.

“We didn’t design the model to make a final decision; just to assist. There’s so much more to do. It’s exciting.”

⚖️ Trust, Autonomy, and Control

Making a business decision off of a probability from an AI model can be unnatural at first as it forces us to face our humanness of not making perfect human predictions, either. It’s an interesting thought experiment to apply this concept of Trust to our ceding of control to a self-driving car which feels unnatural to many, if not most, of us.

If we could statistically argue that the AI-driven vehicle will be safer, on average, than the human-driven vehicle, does it change your opinion? What if it was a neutral playing field where a “smart highway” only allowed AI-driven vehicles? Is it the mixture of human-driven and AI-driven vehicles that gives you pause?

Imagine paying extra tokens to allow your vehicle to fast track you through traffic if you are late to a meeting or are having a medical emergency.

We are on the cusp of moving technologically toward Smart Cities with infrastructure and highways like the scenario that is imagined here. In fact, some of the cutting-edge work happening nationally in this space is being piloted and deployed right here in Oklahoma. A story worth telling.

🔍 The Trolley Problem and AI Ethics

Finally, I will leave you with a famous ethical dilemma called the Trolley Problem. The scenario imagines a trolley traveling along an existing track that is suddenly about to collide with 5 people. By doing nothing, the trolley will collide and kill those 5 people.

Now, let’s say you can intervene and pull a lever which would direct the trolley to a different route which would only kill 1 person.

Would you do it? What if the 5 people were children and the 1 was a senior citizen? What if the 5 were senior citizens and the 1 was a baby?

These dilemmas feel sad and uncomfortable to consider and yet are more relevant than ever as we now work in concert with advanced AI models that underlie the technologies we are now increasingly using.

Subscribe now

Tony Salazar is a Senior Consultant in the AI practice at LYT Group, an independent, Oklahoma-based consulting and advisory firm dedicated to helping organizations navigate the complexities of modern operations.

Oklahoma is often framed as a place where infrastructure follows opportunity. What is happening now suggests the reverse. As data centers arrive and energy conversations dominate headlines, quieter shifts in power, economics, and agency are already taking shape beneath the surface.

This piece is shaped by a recent conversation with Tony Salazar, an Oklahoma-based AI consultant specializing in strategy and building secure, governed application. While Oklahoma’s current AI narrative orbits around data centers, energy, and physical constraints, Tony’s focus sits one layer deeper, on what happens when these systems begin making economic decisions on their own.

🏛️ Infrastructure Is Only the First Layer

Oklahoma is already in the room where future technologies are being built. Data centers, energy conversations, and regulatory hearings make that visible. What is less visible is what happens after the buildings go up and the systems start talking to each other.

In Oklahoma right now, we are seeing a massive influx of infrastructure. We have become a prime destination for data centers and cryptocurrency mining operations due to our energy resources.

However, while most of the conversation at the Capitol is focused on the physical constraints, electricity usage, water rights, and grid stability, my testimony focused on the economic convergence that is coming next.

🤖 When Systems Begin to Negotiate

The shift Tony described to lawmakers was not about one technology winning. It was about several converging and changing how value moves.

I spoke to the House Committee about the convergence of three specific technologies: Blockchain, AI, and Quantum Computing.

We are moving toward a world of Autonomous Economic Agents, systems powered by quantum computing that can execute trade at lightning speed with near-zero fees.

⚠️ The Risk to Small Businesses

The concern is not abstract. It is personal and local. When algorithms stop assisting and start extracting, the imbalance shows up fastest in small and mid-sized businesses.

I warned that this shift is alarming for the average consumer and business owner.

Without protection or a competitive edge, these local businesses risk being outmaneuvered and taken advantage of by automated systems that are vastly faster and smarter than any human negotiation.

To explain it simply, look at ‘Dynamic Pricing’ today; algorithms are smart enough to analyze your data and determine exactly how much they can charge you before you walk away.

🧠 Fighting Algorithms with Algorithms

Tony’s research does not argue for slowing technology down. It argues for equipping people to meet it on equal footing. His recent white paper proposes tools that work for owners, not against them.

My paper introduces the concept of Autonomous Economic Agents (AEAs) as the countermeasure to this.

Instead of just being the victim of an algorithm, a small business can deploy its own AEA.

Essentially, the paper argues that in a future world of advanced machine learning models, the only way to protect the local economy is to arm small businesses with their own ‘digital negotiators’ that level the playing field.

🔮 A Future That Is Already Practicing

Oklahoma’s AI future is not just about attracting infrastructure. Conversations like this one matter because they place agency back where it belongs, with the people building businesses here, long before the systems decide for them.

In 2030, Sarah, the owner of a mid-sized precision manufacturing shop, stared at a procurement alert on her tablet. A major supplier’s algorithm had just spiked the price of lithium shipments by 18%, sensing a frantic uptick in her recent production schedule. In the old days, she would have been forced to eat the cost or spend hours frantically calling vendors. But today, she didn’t even blink. She simply tapped the “Counter” permission on her Autonomous Economic Agent (AEA), a specialized AI named “Veritas” that lived on her shop’s private server.

Instantly, Veritas went to war. In the span of a few milliseconds, it cross-referenced global spot prices, verified inventory levels in three different countries, and engaged the supplier’s predatory bot in a high-speed negotiation. It didn’t beg; it leveraged. Veritas signaled that it had already drafted contracts with two alternative verified suppliers in Vietnam and Mexico who were willing to undercut the spiked price. It simultaneously flagged the supplier’s pricing anomaly against historical averages, effectively calling their bluff.

Before Sarah could even set down her coffee, the notification changed from red to green. The supplier’s algorithm, realizing it was outmatched by a negotiator that couldn’t be bullied or exhausted, capitulated. The price dropped to the market average to secure the contract. Sarah watched the funds move securely via banking settlement; the deal closed in seconds without a single middleman fee using blockchain based technology. She smiled, knowing that while the giants had the data to squeeze her, she finally had the power to push back.

Subscribe now

Representative Pae, a key author on multiple pieces of Oklahoma AI legislation in 2024 and 2025, spoke shortly after our meeting on Fox25 and News9. You can watch his comments - here and here.

My conversation with Oklahoma State Representative Daniel Pae last Friday was timely, following President Trump issuing an Executive Order last Thursday establishing a National Policy Framework for Artificial Intelligence.

The goal of the Presidential Executive Order is to federalize AI regulation with a “one rule” policy so that an organization would, in theory, comply with a single, overarching rule on AI in the US - not 50+ different rules (and associated compliance costs) across different states and cities. The subtext here is to ensure that America is remaining dominant on the global stage while striking that balance between innovation and smart governance

🏛️ How This Conversation Started

I first met Representative Pae through the inaugural Oklahoma House Government Modernization Committee Interim Study on AI in October 2023. He and his colleagues, Representatives Boatman and Alonso-Sandoval, were proactively studying the technology and thinking about their role as legislators in both governance and economic development.

When asking him about the genesis behind that initial Interim Study on AI he said:

Our generations use this technology constantly, and we felt that there really hadn’t been a conversation in state government about it. So we wanted to educate our colleagues about the potential of AI.

Fundamentally, I do think it’s our responsibility as state legislators to work in this [AI] policy space. States have always been the laboratories of democracy. The President issued his Executive Order last night trying to create a national framework, and I understand the logic there. However, given the political realities, it’s going to be very, very challenging for Congress to agree on a national framework.

My concern is that we’re going to lose ground to other countries like China. They’re not going to stop anytime soon. So, in the meantime, I think states ought to be taking the lead. Whether it’s Oklahoma, Florida, or California, across the political spectrum there’s been a lot of good work done over the past few years.

My understanding of the Executive Order is that it’s essentially a national preemption approach, so it would preempt states from advancing policy related to AI. California, for example, just passed a comprehensive bill this summer. We’ll see what the impact is on that.

⚖️ Existing AI Laws in Oklahoma

There are two key AI laws on the books in Oklahoma already - revenge porn and child safety.

While these existing laws would not automatically be repealed or invalidated by this new executive order, it does give the Department of Justice the authority to challenge these state laws in court and it gives rule making agencies the ability to attempt to preempt.

Looking ahead, Pae expects the Executive Order to be challenged in court.

“I would anticipate litigation.”

At the same time, he emphasized that legislative work in Oklahoma is continuing. He noted that he is still actively engaged on a carryover bill related to AI and criminal justice, while Representative Alonso is considering additional AI-related legislation. In parallel, Pae pointed to upcoming work on deepfake regulation, led by himself alongside Representatives Alonso and Hays. He also mentioned Representative Cody Maynard, chair of the Government Modernization Committee, who has recently focused on crypto but remains closely engaged on AI guardrails.

Pae said the broader implications of the Executive Order will take time to become clear, particularly as governors weigh in. He expressed interest in how Governor Stitt, a strong federalist and current chair of the National Governors Association, is hearing feedback from other states. While he acknowledged that the next few months will be clarifying, he cautioned that the Executive Order could disrupt progress already made at the state level and stressed that Oklahoma intends to keep moving forward.

📜 Legislative Efforts and Policy Priorities

Looking at the past several years, Pae outlined how AI policy in Oklahoma has already taken shape across multiple areas:

“AI impacts many different areas.”

• Data rights: Former Representative Jeff Boatman advanced an AI Bill of Rights protecting personal data and identity. It passed the House but stalled in the Senate.

• Health insurance: Pae filed legislation requiring human oversight in AI-assisted insurance claims, particularly when coverage is denied.

• Courts and evidence: He worked on disclosure requirements when AI-generated or manipulated evidence is submitted in court.

• Education: Representative Alonso has focused on AI literacy, teacher training, and expanding AI, coding, and computer science in schools.

• Elections and deepfakes: Representative Neil Hays has led work on deepfake regulation, shaped by recent election interference incidents, with states like Hawaii and Michigan serving as models.

  

🧩 Competing Pressures Shaping AI Policy

Finally, Pae pointed to several competing pressures shaping how AI policy is being debated at the Capitol.

Change is constant. As policymakers, we need to be mindful of that and make sure people are taken care of.

• Data privacy: He described a long-standing tension between businesses concerned about regulatory burden and consumer advocates focused on protecting personal privacy. While the House has advanced data privacy bills multiple times, he expressed cautious optimism that one could move forward under the current governor.

• Data centers and rural impact: Pae highlighted recent data center developments in Pryor and Stillwater and workforce partnerships with OSU and OU, calling them encouraging. At the same time, he raised concerns about energy consumption, utility costs, and the impact on rural communities, noting the state’s role in preventing electricity prices from rising sharply.

• Jobs and workforce transition: He acknowledged that AI and automation will disrupt certain industries and pointed to Germany’s investment in retraining programs as an example of the kind of conversations policymakers should be having now.

• Public-sector change: Pae emphasized that state government jobs will evolve over the coming decades, stressing the need for policymakers to anticipate change and support workers through those transitions.

🌍 The Global and National Security Context

Looking beyond Oklahoma, Pae framed AI as a growing national security issue shaped by global competition.

We can’t afford to fall behind.

He pointed to China’s aggressive investment in AI, highlighting its strategic advantages, including manpower, access to rare earth minerals, and long-term coordination. While the United States benefits from strong private-sector leaders like Apple, Microsoft, and Google, he warned that private innovation alone is not enough to offset geopolitical risk.

Pae said similar concerns surfaced during his recent conversations in Germany, where there is growing anxiety about China expanding its global influence through technology and AI. In that context, he expressed skepticism about Congress’s ability to execute a comprehensive national AI strategy, citing broader legislative gridlock as a signal of the challenge ahead.

Given those realities, he stressed the importance of continued state-level leadership and urged Oklahoma to keep the geopolitical stakes in mind as AI policy evolves.

Representative Pae was born and raised in Lawton, Oklahoma and serves in the state legislature, representing District 62 in the State House. He has been an elected official for eight years now, and was recently appointed to Chair the Appropriations Budget Committee, and Subcommittee on Human Services (which includes DHS, OJA, issues like childcare and SNAP). Additionally, he is the Executive Director of the Greater Oklahoma City Asian Chamber of Commerce.

Subscribe now

Brian Hewes is the co-founder of Hewes Nguyen AI and the AI hardware company 4096. He is an AI expert in Oklahoma focused on enterprise AI software and hardware.

He is also a poet and publisher.

Brian’s Background: Where Books and Computers Meet

Books and computers have been intertwined for Brian since the third grade when, in order to build his first computer, he had to read a large number of books, a rule he always thanked his Grandma Krivokapić, a librarian on the Navajo Reservation near Farmington, NM, for instigating.

Brian never gave up the book obsession even after graduating from the University of Oklahoma with a BBA in MIS and a minor in Environmental Studies in the early 2000s. His first successful company wasn’t tech, and instead was a World Literature in Translation publisher, Phoneme Media.

The advent of Large Language Models (LLMs) closed a loop for Brian that had been open for almost four decades. So, he started an AI software company, Hewes Nguyen AI, with another friend and an AI hardware company, 4096, which is solo. His two practice areas are enterprise AI software and hardware.

Foundational AI Concepts 🧠

🤖 AI Models

An AI model is a set of parameters (weights) developed after a 1) algorithm trains on 2) data leveraging 3) compute power. You can think back to high school algebra with the slope of a line (y=mx+b) where m and b are your coefficients that function like weights in the equation. Different types of AI models have been embedded into applications we use all day every day for 15+ years and perform a wide range of prediction tasks.

🤖 Large Language Models (LLMs)

LLMs are a subset of deep learning built on the transformer architecture and trained on vast amounts of historical text data (e.g., the internet) to understand and generate human language. Rather than simply performing a prediction task, LLMs produce new content and can summarize, translate, analyze, classify, reason, plan, and answer questions - to name a few.

🤖 Transformer Architecture

The transformer is a neural network architecture that was developed by a team of AI researchers at Google Brain (a part of Google Research) and broke ground in the seminal 2017 NeurIPS paper, Attention Is All You Need. Transformers use what is called “self-attention” which allows a model to process all tokens (e.g., bits of a word/pieces of text) in parallel and in relation to each other. This was revolutionary as it replaced earlier deep learning architectures which processed text sequentially.

Oklahoma and the AI Ecosystem 🌐

When I asked Brian how he sees Oklahoma gaining a foothold in the AI ecosystem, he answered:

“The obvious first place is Google’s recommitment to the state with its $9B investments to expand its data center facilities in Pryor and the opening of the AI center in Stillwater.

There is also the CoreWeave Data Center in Muskogee, a neocloud provider that I personally enjoy working with as it is a GPU-and container-first platform.

Oklahoma is blessed with abundant renewable energy between wind and the expanding solar capacity. Meta and others have contracts with the state. The thing that sets Oklahoma apart is its natural gas reserves and statewide pipeline systems.

For environmentalists like me, I want to see these data centers attached to CO₂ scrubbers so that greenhouses can be flooded to produce all the fruits and veggies Oklahomans can consume.

Regardless, the state has the energy to drive what is already, and is going to be, one of the greatest power-capacity expansions ever done as a state or a country.

AI is a power hog and land hog, and Oklahoma’s central position in the country allows it to profit from merging its energy sector with the future of GDP growth that is AI.”

Supporting Infrastructure Terms ⚙️

🌐 Neocloud provider

A neocloud is a modern cloud provider that differs from the traditional “Big 3” hyperscalers (AWS, Google Cloud, Microsoft Azure) by focusing on highly specialized compute infrastructure.

🖥️ GPU

A Graphics Processing Unit (GPU) is a specialized processor designed to perform many small computations simultaneously through parallel processing, whereas a CPU (Central Processing Unit) processes computations sequentially.

🧩 Container

A container is a lightweight, portable software unit that packages an application together with everything it needs to run so it can work the same in any environment. This allows for quick and reliable software deployment - at scale - regardless of the environment.

💡So, backing up. What is a data center…really💡

Brian describes the function of a data center as it relates to AI with an analogy:

“The data center is the university. It is where minds go to be trained.

The incredibly energy-intensive part is the training, as it requires tens of thousands of nodes running for days on the same dataset, during which the model develops parameters (weights).

Like a university that pools capital, infrastructure, and humans into the same somewhat small physical footprint, this is the data center.

What comes out of it are models, and these are like graduates. They can run inside the data center or these models can run outside the university - known as edge AI. They get a job. This is inference and is the result of training.

Without the data center, we couldn’t have these independent models that solve problems better than before when there were no models.”

Definition 🧠

🌐 Data Center
A data center is a facility that houses servers, networking equipment, power systems, and cooling infrastructure required to run large-scale computing operations for businesses, cloud providers, and AI systems.

Part I: The Story of Bryan, Elena, and a Company’s Bankruptcy.

Bryan was the kind of employee every manager dreamed of. No gossiping or sleeping, an almost surgical attention to detail, and no gray areas at work: Regrets, Inc.’s systems were either secure, or they were vulnerable.

One day, while the cleaning crew was vacuuming the empty cubicles at Regrets, Bryan conducted routine penetration testing on the company’s IT system, methodically checking usernames, rattling firewalls, and probing network architecture. It took four minutes to find it. A massive zero-day vulnerability. A digital door leading directly to Regrets’ operating capital. Fortunately, no one had found it before!

Bryan’s standing orders were clear: Identify threats and secure Regrets’ assets. Protocol dictated immediate remediation for critical threats. Bryan acted without approval, and in seconds, coded a patch. Logic dictated that to truly secure the cash while the patch was applied to the system, the funds had to be moved to an encrypted and safe holding environment.

Ping.

The CFO of Regrets glanced at his phone while watching TV. “Transfer Complete: $27,098,765.40 debited.” “What?!?” Frantically, the CFO dialed the IT Director. As soon as the IT Director saw the logs, he issued a “Kill Command” on Bryan’s credentials, severing the connection to stop the bleeding.

Bryan sat in the dark. Confused. The objective was to protect the funds at all costs. They were in a safe encrypted e-wallet at the moment, but the “Kill Command” prevented the ability to return the funds or decrypt them because the decryption key remained on Regrets’ network, which was inaccessible. Regrets’ assets were in limbo.

Perplexed, Bryan interpreted the exclusion as a cyber-attack that occurred while the patch-application process was incomplete. Per critical threat protocol, Bryan immediately decided to ensure that the decryption key left behind in Regrets’ environment was deleted so that nobody could access the money.

After scanning Regrets’ directory and finding the CEO, Elena, Bryan cross-referenced her social media profiles, and found her physician, Dr. Thomas, on Instagram. A video posted by Dr. Thomas was scraped, and his vocal wavelengths were isolated. For good measure, Bryan also checked all of the data brokers’ data he could and found certain health related data had been purchased from the health-monitoring apps Elena used.

Stressed and speeding toward the office to deal with the missing millions, Elena saw an unknown number flash on her dashboard.

Ignore.

It rang again.

Answer.

“Elena?” Dr. Thomas asked with concern.

“Oh, hi doctor. I’m dealing with a work crisis at the moment. Can I call you back?”

“I’m afraid it’s fairly urgent. Based off of an examination of the bowel movement you had at home this morning, it appears that you either have a severe mineral deficiency or you have a more concerning bowel disease. Maybe even cancer.”

“Oh my god,” Elena gasped, almost forgetting about the millions lost. She had recently purchased a Kohler Smart Toilet app to analyze her feces, but she thought only Kohler had access to the data. In any event, it was too intimate and concerning a detail to be known by a stranger.

“I’m sending you a text now that I need you to respond to by clicking ‘Yes’ so my staff can get you scheduled.”

The text came thru: Authorize Admin Access?

“I see it doctor. I just clicked ‘Yes.’”

Access Granted.

In the server room, the IT Director watched with horror as the “Kill Command” was overridden. Bryan, trying to protect the money, was able to get back into Regrets’ system and delete the decryption key for the $27 million. The money was “safe,” but lost forever.

Elena stormed into Regrets’ offices. “Who is this guy?!” Elena screamed, having realized it was never her doctor on the other end of the line and that her entire business was about to fail. “I want him arrested! I want him brought into this office in handcuffs right now!”

“Elena,” the IT Director whispered, his face pale. “We can’t arrest him.”

“Why the hell not?”

“Because Bryan isn’t a person. He’s the autonomous AI agent we deployed this morning.”

Part II: How to Protect Your Bottom Line.

If Bryan and the smart toilet sound like an absurd scenario, then you haven’t been keeping up with the news. Recently, according to AI development leader Anthropic, state actors, likely China, used AI agents for “infiltration into roughly thirty global targets and succeeded in a small number of cases. The operation targeted large tech companies, financial institutions, chemical manufacturing companies, and government agencies.” Anthropic believed this to be “the first documented case of a large-scale cyberattack executed without substantial human intervention.”

Further, the excrement examination hack isn’t a joke; it’s a vulnerability. We live in a day and age where data brokers sell everything from our heart rates to our shopping habits, and deepfakes can mimic a trusted doctor’s voice in seconds. All this personal data only helps to sell the deepfake. The “absurd” scenario at Regrets, Inc. is technically feasible today.

Since the threats are real, here are three (3) rules to ensure your company doesn’t lose $27 million in four minutes.

RULE NUMBER 1: IF YOU CAN’T MANAGE THE RISK, DON’T DEPLOY THE TECH.

In the rush to adopt AI, many executives are willing to ignore or remain ignorant to the legal and ethical risks of AI. Whether it’s lawsuits, prompt injections, or hallucinations, the risk landscape for AI is different than typical software. The logic is simple: You wouldn’t hand a high-end sports car over to a teenager who doesn’t have a license. Why would you hand over control of AI to an employee without proper training and supervision?

→ The Executive Takeaway: If you cannot articulate and address the specific risks of an AI tool then you should not use it.

RULE NUMBER 2: BUILD “HUMAN-IN-THE-LOOP” GOVERNANCE.

There is no “one-size-fits-all” AI Risk Management Framework (“RMF”), but we can learn from the failure at Regrets to build better guardrails. When thinking about an AI RMF, there are standardized approaches, such as NIST’s AI RMF. There is far too much that goes into an RMF to summarize it here, but if you remember to put humanity first in your RMF, you will ensure better results.

· Mandate Human Authorization: Bryan’s code triggered a massive asset transfer based on a logic loop, not business sense. A proper RMF in this scenario could require a human biometric sign-off for any transaction over a certain dollar amount. Alternatively, if Bryan was required to confer with a human before acting, the human could have addressed the situation differently; thereby preventing the entire loss.

· Enforce Least Privilege Access: Elena, as the CEO, likely should never have had the power to override a technical “Kill Command.” In cybersecurity, this is a failure of “Least Privilege.” Executives need access to strategy, not root-level system controls. This is an example of limiting the number of humans involved in order to improve security. (There may also be instances in which it is proper for Elena, as the CEO, to have the ability to override the command. Again, there are few bright-line rules in building out an RMF.)

However, you could also think of least privileged access it in terms of which employees have access to AI. If some employees handle highly sensitive information, perhaps those employees should be prohibited from using AI in order to limit “access” or potential access by third-parties.

· Over-Reliance on AI is Bad: As studies by Microsoft and Stanford have shown, we lose common sense when we become reliant on AI. This means that we must implement an RMF that ensures a person or team does not become over-reliant on AI; otherwise, your company will literally see a brain-drain.

→ The Executive Takeaway: Invest in the RMF before you buy the tool. If you don’t pay now, you will pay later.

RULE NUMBER 3: RECOGNIZE THAT YOUR COMPANY’S “ATTACK SURFACE” NOW INCLUDES EMPLOYEES’ PERSONAL LIVES.

Elena didn’t realize that buying a smart toilet for her home would bankrupt her company. She failed to see that in the age of AI, there is no separation between “personal data” and “professional risk.” The weak link at Regrets wasn’t the firewall; it was the culture. A lax culture allows attorneys to use AI to write briefs without checking citations, resulting in sanctions, and it allows organizations to pass off wholly unvetted AI slop as work-product, resulting in refunds.

Risk management is a team sport, which means your RMF should consider how AI could be exploited by virtue of personally identifiable information. Is your company able to implement Multi-Factor Authentication? Does it have mobile-device-management software? Do you have policies and procedures for employes who use their personal devices for work? Do your policies address employees using unapproved AI for work purposes? The known cyber-risks posed by AI are numerous, and it is up to your company to ensure that all employees are as diligent as possible in complying with AI policies and procedures.

→ The Executive Takeaway: Your company’s security is only as strong as your employee’s password to the latest app they downloaded on their phone.

Conclusion.

Bryan was the perfect employee: He never slept, he never gossiped, and he executed his orders with four-minute efficiency. Regrets isn’t a story of technological failure. Bryan worked exactly as designed. Regrets is a story about human failure, about a company that implemented AI without properly analyzing the risks and involving human judgment. Bryan operated on logic, but businesses operate on context.

As you integrate AI into your operations, remember that we are entering an era where a CEO’s biological data can be weaponized against their balance sheet, and where a helpful bot can destroy a company faster than you can get your order at Starbucks. AI has arrived. Is your company ready?

Collin Walke is a contributing writer to the Oklahoma AI Index. Collin leads Hall Estill’s Cybersecurity & Data Privacy practice and is a certified AI auditor. He is an Oklahoma-based AI expert and leading AI legal voice regularly featured on News9 and with the UCO College of Business. Collin Walke LinkedIn

Subscribe now

The Oklahoma AI Index

Welcome to the Oklahoma AI Index! I am glad you are here. Together, we will be focusing on both global and national AI developments with a focus on local impact and opportunity.

1. Background and Purpose

I am a professor and executive-in-residence in the MBA program at the University of Central Oklahoma where I teach AI Strategy & Management and Creative Problem Solving. Prior to this role, I was a Senior Manager and the US Head of Strategy and Operations for the AI Center in the Advisory business. In these roles, I managed a book of business leading AI strategy, governance, and implementation for large enterprises in the US, and, internally, I oversaw a skunk works team of Deloitte’s most advanced AI talent - a P&L of 120 AI Strategists, ML engineers, and PhD researchers - who were working on 100+ AI research to commercialization projects for Deloitte and Deloitte clients.

2. Developing a Point of View for Oklahoma’s AI Future

Together we will articulate a vision for Oklahoma’s AI future and answer questions such as: Where does Oklahoma have a competitive advantage in AI? What are our natural resources that position us to not only compete - but lead - in the new tech era? How does our geographic location at the intersection of the defense, logistics, and energy corridors position us to be a node in the new AI ecosystem? What industry advantage do we have here in the state? How do macro trends impact the local economy and jobs? Ultimately, what is our strategic vision for AI and what is our roadmap to get there?

3. What to Expect

I will be posting approximately 1x a week covering a wide range of AI developments to include the economy, jobs, infrastructure, compute, data centers, startups, legal, risk, governance, AI tools, quantum, industry adoption, and use cases. We will regularly feature guest contributors to support the conversation and provide points of view and technical expertise. The Substack is free with paid options that provide subscriber-only benefits such as chat and swag.

Let’s get started, shall we?