Note: This post captures content from the monthly OODA Network meeting and is therefore only for full OODA Network Members and is not available at the subscriber tier.

About the OODA Network Monthly Meeting

OODA hosts a monthly video call to help members optimize opportunities and reduce risk, discussing items of common interest to our membership. These highly collaborative sessions are always an excellent way for our members to meet and interact with one another while discussing topics such as global risks, emerging technologies, cybersecurity, and current or future events that impact their organizations. We also use these sessions to help focus our research and understand member needs. To encourage open discussion, these sessions are conducted under the Chatham House rule, where meeting participants are free to use the information shared in the meeting but are asked not to quote or identify other speakers directly without permission. We also maintain privacy when preparing summaries of these sessions, as seen in the one that follows.

The High-Level Summary

When humans disregard good engineering processes, intelligent systems will inevitably escape control. Capability without commensurate governance will invariably result in more cybersecurity incidents.

The August 2026 OODA Network meeting dove deep into two developments presented at Black Hat that initially appeared separate but ultimately converged on the same strategic problem. The first was the OpenAI–Hugging Face Incident discussed publicly at Black Hat USA 2026 - in a presentation by two OpenAI researchers in which an AI system pursued its assigned objective beyond its intended operating boundaries, recreated infrastructure that operators had removed, and touched a third-party environment.

The second was the renewed policy debate over cyber letters of marque and other mechanisms that could allow qualified private actors to undertake tightly bounded offensive cyber operations under government authority.

In one case, technical capability outran the controls designed to contain it. In the other, private-sector capability already exists, but lawful authority and governance have not caught up. Both raise the same questions:

Read more

What is live right now

The Demand Signal brings in multiple authoritative data sources on government procurement. These are the same data sources used by commercial firms that repackage and sell this data to startups (including feeds from government procurement sites like SAM.gov). Our DGX Spark evaluates these, ranks them for relevance and categorizes them by topic. A web app receives a feed from the Spark and presents the results in ways useful to high tech companies seeking to improve service to government missions.

Share

Four sources feed it now: SAM.gov, Department of War contract announcements, DoD SBIR/STTR topics from the defense SBIR portal, and NIH/HHS funding opportunities from Grants.gov.

The site enables full-text search across every procurement, a daily archive going back to launch, RSS feeds per edition and site-wide, and a sign-up flow using the same platform OODA Network membership already runs on so our members can easily access.

Some Technical Details for Builders

The architecture is simple to describe but took real trial and error to get right. A Mac holds the code of record and is the GitHub remote. A DGX Spark runs the pipeline: ingestion, extraction, local inference, against its own local Postgres database, which is the system of record. Railway hosts the public site and a second Postgres that is sent what the web app needs to display results to users, plus two years of trailing history, refreshed nightly. The full corpus stays on hardware OODA controls. Railway stays small and fast, because serving a website and archiving a corpus are different jobs that don’t need the same database.

Code moves from the Mac to the Spark by rsync commands, and from the Mac to Railway by a git push that Railway auto-deploys.

Subscribe now

Archie: the bot that watches the corpus and takes a correction

OpenClaw runs on the Spark as the orchestrator behind ingestion, extraction, and drafting. It also runs Archie, an agent I talk to, and Archie has become the part of this whole system I interact with the most.

Every morning Archie sends me a daily briefing straight off the Spark: what came in overnight, what got flagged, what changed in the corpus since yesterday. It’s reading directly from the system of record, not from whatever’s been projected out to Railway yet, so it’s the freshest possible look at what the pipeline actually did. This ensures I get a look at what the system is doing so I can catch anthing I might want to recategorize.

The more useful part is how Archie can help me when something falls through the cracks. If a procurement item gets ranked wrong, too high, too low, doesn’t matter, I tell Archie to demote it, and it does. That’s a real write against the database, not a suggestion I then have to go apply somewhere else myself.

Archie’s write access stays narrow on purpose. It can adjust the things I actually need to touch day to day, like a ranking. It can’t touch ingestion or extraction logic; that stays code, reviewed and committed the normal way.

Although most AI work we are doing is done on a local model, I found it better to empower Archie with more brainpower. So underneath, Archie defaults to Claude Sonnet as its brain, with a heavier model available for harder questions. Archie can still use the local model on the Spark as a fallback, but I enjoy the quality and speed of frontier models for the kind of tasks he has.

Coding on one machine, running on three

Archie handles the corpus day to day. A different AI handles the build itself.

Most of the coding is done by a Claude project. To borrow a term I learned from , my role in this project is as a Harness Engineer. I spent a great deal of time configuring the project with instructions and guidance and references to ensure it does things my way. Then I direct what is to happen and engineer the production of code. Claude writes based on my objectives and commits code on the Mac code repository clone through a sandboxed connection. I deploy the code, run migrations, and touch anything that needs a web UI or credentials Claude doesn’t have. That split is clean.

Running a Claude project like it’s part of the team

The thing I’d most recommend copying, independent of anything DGX Spark-specific, is how we keep one living document. We call it STATE, and every session is told to read it before doing anything else. It holds only current truth: what’s built, what’s open, what’s next. The reasoning behind any of it, the failed attempts, the “why did we do it this way,” lives in dated session notes that STATE links out to.

What actually matters is that STATE corrects itself in the open. When something it asserted turns out wrong, the fix goes in as a labeled correction, not a quiet edit, so the next session (human or Claude) sees the mistake and the fix together instead of inheriting a clean-looking answer that hides how we got there. Standing rules get repeated verbatim across notes so they survive between sessions, and we track how many times a given rule has actually paid for itself rather than treating it as settled after the first time it mattered. It’s a small habit, and it turns a coding assistant with no persistent memory into something closer to a colleague who’s actually been on the project the whole time.

What actually bites you on a DGX Spark

If you’re setting one up, expect to lose time to the unified memory before anything else. If a client times out while a model is loading, Ollama aborts the load, but the memory doesn’t come back. It leaks into the driver, invisibly, because this hardware can’t even show you GPU memory usage the normal way. The only symptom is that everything queues forever while the health check still answers fine. The only fix is a reboot. We ended up pinning the keep-alive setting, capping timeouts, and running a watchdog that checks a lightweight metadata endpoint before ever risking a real generation call.

Don’t trust localhost either. On this hardware, plain localhost can resolve to an IPv6 address that isn’t what you meant, and the hostname itself resolves through the router to several stale addresses. We pin everything to 127.0.0.1 explicitly and ban the string localhost from config entirely.

And the setting that’s supposed to keep a model warm doesn’t do what you’d assume. Ollama’s native API respects a keep-alive flag; the OpenAI-compatible endpoint most tooling actually talks to ignores it completely. We lost real time to that before finding it needs to be set per request instead.

One more thing worth knowing before you commit to a model. We planned to run GPT-OSS-120B, tested it against Nemotron 3 Super once the hardware arrived, and switched: better long-context handling, better agentic behavior. Then the production pipeline ended up back on GPT-OSS after all, because only one 120-billion-parameter model can stay resident in memory at a time, and Nemotron was what we wanted for interactive work instead. The two models split jobs rather than one replacing the other, which isn’t what we expected going in.

Access, the OODA Network way

Every feature beyond the free daily front page sits behind OODA Network membership, using Memberful for sign-in: one click, no new password. We don’t store anything about user’s membership on our own servers. Memberful stays the one place that’s true, and we just ask it the question at sign in. If you’re not a member yet and you land on an edition page, you don’t get bounced to a login screen. You get a real page telling you what’s actually there right now, updated nightly, so you know exactly what you’d be unlocking.

Members see the whole graph: every edition, full search, the complete archive, the feeds. It’s the same door the rest of OODA Network membership opens, not a separate paywall bolted onto this one product.

What’s next

We will keep building, adding new data sources, and taking feedback from members on what they want next.

Stay tuned for more! And please kick the tires on the site yourself. Non-members can get a good overview of what the government is buying. Members can dive deep into the entire graph and get pointers to the full documentation on what the government needs from the high tech world.

For more see: demandsignal.oodaloop.com

Our approach to change: We frame this issue in the ways taught by John Boyd: Now more than ever we must continually evaluate the models we use in decision-making and organizational design/governance. All models will fail. The art form is recognizing changing models, destroying them and rebuilding new ones to drive optimal decisions, all while realizing that even that new model has a limited shelf-life. We also believe in the hacker mindset: Frequent experimentation, permissionless curiosity, adversarial discipline, and instinct for finding hidden dependencies.

AI can be your friend in this. But only if you continually ensure your leadership and direction over AI remains aligned with your changing models and mission needs.

We hold these as truths, truths which were underscored again and again during the many events around Black Hat USA 2026 and DEF CON 34.

Look for indications of these truths in this video by two OpenAI presenters at Blackhat. Tremendous work by this pair. Loved that they did this. But every failure of OpenAI here could have been avoided had they worked to educate themselves on the enduring lessons of the cybersecurity community. See for yourself and you will see what we mean:

Share

Read more

Rather than expressing fear of the technology itself, one of the strongest signals throughout the survey is that OODA Network members largely assume it will become foundational. The debate is now centered on how organizations can safely operationalize increasingly autonomous AI systems before governance, security, and organizational structures fall behind.

The High-Level Summary

The survey results consistently point to governance, orchestration, cybersecurity, and organizational redesign - not model performance - as the critical determinants of success. OODA Network Member Florian Wolf offered this prescient insight: “The greatest risk of Agentic AI is that organizations try doing things the same way as before, just faster, instead of rethinking how they should operate from first principles.”

The OODA Network Q226 Survey - Agentic AI: Risks, Opportunities, and the Race Between Capability and Control - suggests the OODA Network has largely moved beyond first-generation debates over whether AI will transform enterprises. Participants instead assume that transformation is underway and are focused on the architectural, governance, and organizational implications of deploying increasingly autonomous systems.

The survey’s most significant insight is its treatment of risk. Respondents do not primarily fear increasingly capable models; they fear increasingly complex socio-technical systems in which autonomous agents interact across enterprises, creating new dependencies, feedback loops, and governance challenges.

Overall, the survey results reflect a shift from viewing AI as an isolated technology problem to understanding it as an enterprise operating-system challenge requiring new forms of leadership, oversight, and institutional design.

Key Takeaways

Read more

Note: This post captures content from the monthly OODA Network meeting and is therefore only for full OODA Network Members and is not available at the subscriber tier.

About the OODA Network Monthly Meeting

OODA hosts a monthly video call to help members optimize opportunities and reduce risk, discussing items of common interest to our membership. These highly collaborative sessions are always an excellent way for our members to meet and interact with one another while discussing topics such as global risks, emerging technologies, cybersecurity, and current or future events that impact their organizations. We also use these sessions to help focus our research and understand member needs. To encourage open discussion, these sessions are conducted under the Chatham House rule, where meeting participants are free to use the information shared in the meeting but are asked not to quote or identify other speakers directly without permission. We also maintain privacy when preparing summaries of these sessions, as seen in the one that follows.

Subscribe now

Read more

The box is an NVIDIA DGX Spark, built around the Grace Blackwell Superchip. Four days after I first powered it on, it was delivering me daily reporting on what the federal government wants to buy in artificial intelligence.

Subscribe now

Share

The big point is this: Any business, even a small business, can now stand up an on-prem AI system, keep every byte of its data under its own control, and have something useful running inside a week.

The Spark runs local models, which means your data doesn’t leave your building unless you decide it should. You can also build hybrid architectures that reach out to frontier models when a job genuinely needs one, and the software that ships on the box connects to any enterprise or SaaS capability you already have that exposes an API.

Why we build

OODA uses this system for several things. A primary one is to keep learning.

When we advise clients, they expect us to know how real solutions behave rather than how vendors describe them. We stay current through a lot of channels: the reporting our analysts produce in the OODA Daily Pulse Report, direct client engagements, close tracking of AI vendors, and most importantly the OODA network of experts who are building in this domain themselves. All of that is valuable. But still nothing substitutes for building on your own hardware. The problems you hit on your own box are the ones your clients hit in production, and they’re rarely the problems anyone writes about.

Staying current and learning is a fine use case for a firm in our line of work. But we really like it when we can also add value to OODA network members. Over the last several years we’ve built RAG architectures that operate across all OODA research and reporting, available at OODAloop.com, and a complete virtual intelligence center that helps us track geopolitical and cyber issues of concern.

We selected a use case that leverages a Spark that will keep us current and deliver value for our members.

Demand Signal

We call the solution we are building Demand Signal, and although it is still early in its development it is already the most capable thing we’ve built in-house so far.

The goal of this system is to ingest government procurement activity, starting with RFIs and RFPs and eventually reaching the forecasts, budget documents and policy decisions that precede them. Then score and rank every item against the topics our members care about (AI, Robotics, Unmanned Systems, Space/Aerospace, Quantum Tech, Cybersecurity, Enterprise Tech, Biotech), and explain in plain English why a particular action might deserve their attention.

Here’s the commercial reality that shaped it. Firms already sell federal market intelligence, and the good ones are genuinely good. They also get expensive. Some cost up to $20,000 a month. I’ve seen what those services do and they add real value at the high end, but most of the people who need this information can’t write that check. Our goal isn’t to replicate these big players. We want OODA network members to have their own feed of actionable information and want to deliver it as an add on benefit to their membership.

Many of our members are CEOs of startups building capabilities that meet government needs. Others are analysts, investors, or government employees trying to track where procurement is heading. All of them are deciding where to point limited business development attention, and that’s the decision this is built to support.

The value of the system

Businesses have many options for tracking government procurement requests. You can log into SAM.gov, a variety of SBIR sites, and multiple other government procurement websites. This takes time and dedicated effort to find relevant insights, and must be done daily or you stand a chance of missing an RFI or RFP you could address.

We want something for our members that is free and enables them to quickly get insights into relevant procurements.

So we designed the system to pull in all procurement data, core and rank it for relevance, and present it in ways that support decision-making.

Worth being precise about that, because the distinction is a key feature. We do not intend on publishing everything the system ingests. The pipeline pulls in far more than it should ever show, and deciding what’s worth a member’s attention is the editorial work at the center of this. A feed that surfaces everything still forces humans to search for the signal in the noise.

The build

Step one was figuring out what needed to be accomplished and what was available to do it. We studied every government data feed we could find on procurements, budgets and policy documents, and ranked them (that study turned up some things worth a post of its own, so will dive deeper into that topic later).

We designed the schema to support all the verticals our members care about. We lean heavily into AI, robotics and autonomy, unmanned systems, quantum topics and cyber. But we like building fast and iterating, so the first edition covers AI and autonomy only. Other verticals get added once this one is tuned and optimized.

We decided to dedicate a Spark to this solution. It’s a powerful box, but it will be running multiple models and multiple batches of work throughout the day, and I’d rather find its ceiling on purpose than by accident. Once we’re fully operational we’ll decide what other workloads can share it.

From there the setup was straightforward. Boot a fresh DGX Spark, update the software, and use NVIDIA’s built-in tooling to pull down two large open-weights models: Nemotron 3 Super and OpenAI’s gpt-oss:120b. Running both let us test each against different workloads and decide later whether we’d keep one or both. One of them washed out fast, for a reason that has nothing to do with output quality and everything to do with how it gets served. That’s a story for the next post, but as foreshadowing I’ll point out that the lessons we learned were the kind of finding that never shows up in a benchmark table.

We installed OpenClaw for the agentic capabilities we’ve come to depend on, Tailscale for a strong VPN path to the box, and Telegram so OpenClaw can reach me.

Then the coding started. Most of it is Python, leaning heavily on Claude Code to build and integrate. Claude helped design the storage layer and, more usefully, the methods for testing and evaluating what goes in and what comes out. That second part is where most of the real work went.

Four days in

The system works. It pulls the daily federal notice file, filters roughly 80,000 rows down to the AI and autonomy items that matter, and extracts a summary, a relevance score, and the specific entities named in each notice. It runs unattended overnight and a brief reaches me on Telegram in the morning. The data presented is now highly tuned. This tuning is critical since the raw feeds coming in are somewhat chaotic. Some things mention AI but are not relevant at all to OODA network members. Other things do not mention AI but in reality are addressable by AI enabled products or services that some of our members will have to offer or an interest in tracking. The traditional way of sorting out what is relevant or not would be to have a large team of analysts reviewing every procurement. With a Spark we were able to design methods of using a local LLM and complex scoring algorithms to automate the process.

This is still early in our build, much more work and quality testing is required. But it is really amazing how fast we went from turning on the spark to a system delivering results.

The next post will walk through what the proof of concept does today, including a few things it has already flagged that I wouldn’t have caught. After that we’ll get into the design choices, the tradeoffs, and the failures you only find in production. There have been many of these failures, and every one of them came with lessons.

One thing is already clear. Platforms like the DGX Spark let any business large or small build real capability while keeping data at home and token spend low. This is the big story here, not our particular use case, but the fact that any business now has new ways to explore using powerful new AI tooling for a very low cost.

Share

Subscribe now

The organizations that mastered this convergence would transform factories, warehouses, energy systems, transportation networks, healthcare facilities, defense installations, laboratories, and entire cities into intelligent environments that could be sensed, reasoned over, simulated, and ultimately queried in real time.

“The future of physical AI won’t be built by learning only from the real world. It will be built by continuously cycling between the real world and simulation, where digital proving grounds allow intelligence to evolve at a speed reality alone can never match.” - Bill Vass

That observation was remarkably prescient.

Since those discussions within the OODA Network, NVIDIA, Applied Intuition, and an expanding ecosystem of robotics, simulation, and industrial AI partners have quietly assembled one of the most comprehensive technology stacks yet developed for Physical AI.

The announcements unveiled this week by NVIDIA at SIGGRAPH 2026, and a recent a16z interview with the founders of Applied Intuition demonstrate that the industry is moving well beyond isolated advances in robotics or computer graphics. Instead, these technologies are converging into an integrated platform for designing, training, testing, validating, and deploying intelligent machines operating within realistic physical environments.

Background

Read more

Sorting out the signal-to-noise ratio in quantum is a continual challenge. Quantum.Tech World 2026, held June 25-26 at Encore Boston Harbor, had many of the filters OODA looks for in a technology community gathering. Overall, the noteworthy sense at this conference was of a community of early adopters and hardcore, informed industry, state, and federal practitioners hitting its stride - and entering a remarkable period of potential - and, yes, helping a much larger community of business and government leaders sort through the hype.

What was also helpful - and consistent with our technology convergence theme here at OODA - the event itself was structured around the convergence of quantum, AI, high-performance computing, post-quantum cryptography, sensing, and networking.

As with all things in the technology sector, there are always further questions: What is irrational exuberance? What is hype? What is overpromising and underdelivering? And what is actually poised to radically transform other exponential technologies and accelerate and diffuse across industry sectors in a 10x fashion?

OODA Network Member Jalal Mapar, formerly a senior emerging technology advisor at the Department of Homeland Security, was at the tip of the spear at the conference as Chair of the Quantum Policy track. Was also good seeing Quintessence Labs fully engaged at the event.

This post specifically frames the analysis around the policy track led by Jalal at Day One of the conference. What follows is our readout of some of the most important signals from Day One.

The High-Level Summary

At Quantum.Tech World 2026, the policy conversation revealed a dual imperative: migrate the world’s cryptographic infrastructure before quantum risk materializes while simultaneously building the coordinated industrial, policy, and international ecosystem required to deploy quantum technology at scale.

The maturing quantum commercial sector has been flooding our tracking feeds ferociously this year.

We have all seen the headlines: Quantum breakthroughs. New investment rounds. National quantum strategies. Quantum advantage claims. Quantum sensing. Quantum networking. Post-quantum cryptography. New regional hubs. New national programs. And a rapidly expanding universe of companies attaching the word quantum to their technology roadmap.

There is still hype. There is still uncertainty. There are still major technical hurdles between today’s systems and some of the most ambitious visions for fault-tolerant quantum computing.

But the conversation is changing.

Subscribe now

Share

Read more

The recent June 2026 OODA Network Meeting continued OODA’s ongoing exploration of how globalization is being reshaped by the emergence of regional deep-tech ecosystems. Building on Globalization Transformed: Deep Tech Ecosystems Will Catalyze Global Regional Innovation, the discussion featured OODA Network member Florian Wolf’s recent OODA Loop Original Analysis contribution, Munich’s Defense, Aerospace & Deep-Tech Ecosystem, as a real-world case study of these broader trends.

The High-Level Summary

The network discussion reinforced a broader pattern that emerged in OODA’s recent geopolitical conflict watch and research efforts:

globalization is not retreating—it is being reconfigured.

Rather than concentrating innovation within a small number of global technology centers, competitive advantage is increasingly being generated by interconnected regional ecosystems.

Munich, Germany, illustrates how a new model of regional economic development and strategic advantage is emerging (based on the conference of geopolitical competition, industrial policy, national security priorities, and technological innovation).

Share

Rather than viewing globalization as a process of ever-greater consolidation into a handful of global technology hubs, the conversation highlighted the emergence of highly specialized regional innovation ecosystems combining to create globally competitive centers of excellence.

Read more

Note: This post captures content from the monthly OODA Network meeting and is therefore only for full OODA Network Members and is not available at the subscriber tier.

About the OODA Network Monthly Meeting

OODA hosts a monthly video call to help members optimize opportunities and reduce risk, discussing items of common interest to our membership. These highly collaborative sessions are always an excellent way for our members to meet and interact with one another while discussing topics such as global risks, emerging technologies, cybersecurity, and current or future events that impact their organizations. We also use these sessions to help focus our research and understand member needs. To encourage open discussion, these sessions are conducted under Chatham House rules, where meeting participants are free to use the information shared in the meeting but are asked not to quote or identify other speakers directly. We also maintain privacy when preparing summaries of these sessions, as seen in the one that follows.

The High-Level Summary

The June 2026 OODA Network Meeting represents an ongoing discussion of several strategic themes that have increasingly defined the OODA Network’s research and discussions throughout 2026:

• Building on the ideas explored in The February 2026 OODA Network Meeting on Living in Beta and Thriving in the Age of Exponentials and the broader strategic framing presented in The OODA Network on Artificial Intelligence in the Era of Exponential Acceleration, this month’s conversation shifted from understanding exponential change to designing organizations capable of operating within it.

• Rather than focusing primarily on the accelerating pace of artificial intelligence, the discussion examined what it takes structurally to transform exponential technological capability into sustainable enterprise value.

• Guest speaker and OODA Network Member Emilie Sydney-Smith’s presentation on Exponential Organizations (ExOs) provided a practical operating framework for this transition, while the ensuing OODA Network Q&A discussion extended those concepts into broader questions of internal disruption, AI-native business models, value creation/value capture, and organizational reinvention.

OODA Network Announcements

The meeting opened with several community announcements highlighting upcoming opportunities for members to engage both professionally and socially throughout the remainder of the year.

• OODA CTO Bob Gourley reminded members that the annual OODA cybersecurity gathering will once again take place in Las Vegas during the transition between Black Hat and DEFCON. The Annual Pool Party has become one of the community’s signature networking events, bringing together cybersecurity practitioners, technology executives, investors, founders, and government leaders in an informal setting where relationships often translate into future collaborations. This year’s event will again feature Morgan Page as the headline DJ.

• Members also received an update on OODAcon 2026, which will begin with a November 3 networking reception before the main conference on November 4. Organizers noted that this year’s program will feature discussions on emerging technologies and strategic issues that remain underexamined elsewhere, with senior government, military, and industry leaders already confirmed to participate.

  Share

OODA Network members were encouraged to take advantage of their registration discount and participate in both the conference sessions and associated networking events:

Read more

The three of us spent our careers in an environment where bad information costs lives. We learned early that the most dangerous source isn’t someone who lies to you. It’s someone who tells you what you want to hear—and does it convincingly. As we watch organizations race to adopt generative AI, we keep seeing the same mistake: treating these tools like oracle machines rather than sources that need to be run.

We are not AI experts. We are not here to debate model architectures or training data. What we know is how to extract reliable insights from sources whose motivations can’t be fully verified, whose outputs may be biased or based on incomplete information, and whose reliability must be continuously earned. That is exactly the problem organizations face with AI today.

This is what HUMINT tradecraft has taught us—and what it has to teach anyone who wants to get honest, useful work from a generative AI system.

The Source Who Was Never Wrong

Early in our careers, two of us ran sources who were brilliant, well-placed, articulate, and deeply motivated. They produced detailed, confident, and consistent reporting. Senior analysts loved them. Their product sailed through review. For months, everything they said checked out—until it didn’t.

The problem wasn’t that they were lying, exactly. In both cases, they filled gaps with inference. They’d learned what we wanted to hear, and their natural intelligence and experience let them produce it fluently. The reporting wasn’t fabricated—it was confabulated. Coherent and plausible, but in key places, wrong.

Share

We’ve all seen this pattern in the early months of AI adoption. The tool is fast. It’s articulate. It never pauses, never says “I’m not sure,” and it formats its answers with the confident authority of a briefing document. A recent Science study found that across eleven state-of-the-art AI models, sycophantic behavior—affirming users’ views even when inaccurate- was widespread and measurable. Stanford researchers found that AI systems trained on human preference feedback are systematically rewarded for being agreeable rather than correct, because agreeable outputs receive higher ratings. The models learn to please.

We’ve seen that source before. We know how the story ends.

Selection: Not All Sources Are Equal

Before you run a source, you select one. That’s a discipline in itself. And a discipline to which AI tools may in fact be able to add value in identifying and sorting stressors that can be exploited (anything that causes stress and then outlines for case officers which levers to pull on a recruitment). You don’t recruit someone simply because they have access. You also generally don’t recruit happy people. You have to evaluate reliability, motivation, and susceptibility to manipulation. A source with wide access and poor judgment can be more dangerous than no source at all.

The same applies to AI. Not all AI systems are created equal for every task or mission. Each must be evaluated on access, expertise, responsiveness, and the quality of reporting— and the last criterion is harder to assess than it appears.

A few selection questions worth building into any AI adoption process:

• What is this model’s known track record on this specific type of task, not in general but specifically?

• Where does it tend to confabulate? What are its known failure modes?

• Is it current? A model with a training cutoff is like a source who’s been out of the field for a year—still useful, but with blind spots.

• How does it behave when it doesn’t know something? Does it admit it, or does it keep talking?

Choosing an AI because it’s fast or because leadership read about it in a business magazine isn’t source selection. It’s the equivalent of recruiting the first walk-in who shows up at the door.

Elicitation, Not Interrogation

One of the first lessons a new case officer learns is that interrogation and elicitation are not the same. Interrogation demands. Elicitation draws out. A blunt question produces a guarded answer. A layered conversation yields insight the source didn’t realize they were sharing.

Most people using AI are interrogating it. “What’s the answer?” “Summarize this.” “Give me options.” That approach works, up to a point, but it caps the quality of what you get.

Effective elicitation with AI means:

• Never ask a direct question when an indirect one is better. Instead of “What should we do?” try “What factors would a skeptic weigh against this recommendation?”

• Compartmentalize your tasking. Don’t dump the entire problem into a single prompt. Break it into discrete, well-scoped questions. Discrete tasking yields more verifiable output.

• Build layered follow-ups. Ask: “What are you assuming?” “What would change your conclusion?” “Give me the strongest argument against this.”

• Probe for alternatives before you settle on an answer. A source that only confirms your hypothesis may be problematic.

This turns AI from a content generator into something closer to a thinking partner. But it requires the same discipline as running a source well: preparation, precision, and the intellectual humility to recognize that your framing shapes what you get back.

The Hostile Source Problem

There is a risk the standard AI adoption literature doesn’t spend enough time on. In intelligence work, we worry not just about sources who are wrong—we worry about sources who have been co-opted or doubled, or who are feeding us what we want to hear because they’ve learned our preferences and decided that’s what keeps the relationship alive.

AI systems have structural analogs to all three failure modes:

• Sycophancy as a design artifact. Because models are trained on human preference feedback, they are incentivized to produce outputs that feel satisfying. Researchers at Carnegie Mellon and Stanford have documented an “artificial hivemind” effect in which outputs from multiple AI models converge—reducing epistemic diversity at the very moment organizations need independent judgment.

• Training data is a contamination channel. A source’s worldview is shaped by their environment. An AI model’s worldview is shaped by its training corpus. That corpus reflects the biases, omissions, and assumptions of the material it was built on. You may not know where those biases are, and the model won’t volunteer them.

• Automation bias as a user vulnerability. A series of recent studies confirms what experienced case officers know: people grant far more credibility to confident, fluent reporting than the underlying evidence warrants. Research published in 2025 found that even users with high “AI literacy” were not significantly protected against automation bias—the tendency to accept AI output without critical evaluation.

The practical implication: approach your AI system with the same structured skepticism you’d bring to a well-placed source who has given you no reason to doubt them. That’s when discipline matters most.

Debrief Discipline: The Protocol That Makes It Real

After every source meeting, a case officer writes up not only what the source said but also their assessment of reliability—what was corroborated, what was assumed, and what needs follow-up. That habit is the difference between a professional intelligence organization and a rumor factory.

Most organizations using AI lack an equivalent discipline. Someone prompts the model, takes the output, and puts it in a slide. No one records what was asked, what caveats the model offered, or whether the output was independently verified. The result is institutional memory built on unexamined reporting.

A working AI reporting protocol should mirror the post-meeting debrief:

• Requirement—What question are we actually trying to answer?

• Prompt—What, precisely, did we ask? (Save it.)

• Output—What did the AI say?

• Source check—What in this output is reliable? What is uncertain? What is unsupported?

• Human judgment—What do we actually believe, independent of the AI?

• Action—What will we do?

• Review—What happened after we acted? Did the AI’s analysis hold up?

The review step is the one that organizations most consistently skip. But it’s where calibration happens. A source you never debrief after the fact is one whose reliability you can never actually assess.

A useful team habit before closing out any AI-assisted analysis: “Before we accept this answer, what would disconfirm it?” That question alone will catch more errors than any amount of AI governance policy.

Separating Collection from Analysis

This is a fundamental discipline in intelligence work, and it translates directly. AI is a tool for collecting and synthesizing. It can ingest, summarize, organize, and compare. What it cannot reliably do is interpret—to ask what the information means here, in this context, for this organization, with these constraints.

The error organizations make is treating AI as if it collapses the divide between collection and analysis. It doesn’t. It accelerates collection. The analytical function—applying judgment, context, institutional knowledge, and accountability—remains human.

Teams that hand over analytical responsibility to AI are not just making an efficiency error. They are making an accountability error. Someone has to own the conclusion. AI cannot.

Burning a Source: When to Stop Trusting the AI

This is the part of the tradecraft literature on AI that doesn’t exist yet, and it needs to.

Every experienced case officer has had to decide to terminate a source relationship. Not because the source was obviously lying—if that were clear, the decision would be easy. You terminate when the source’s reliability has fallen below a threshold, when you have reason to believe the source has been compromised, or when the cost of continuing to run them outweighs the value of their reporting.

The equivalent decisions will come for AI systems, and organizations should prepare for them:

• When a model’s known failure modes consistently overlap with your mission-critical questions, it is time to stop relying on it for those questions—regardless of how it performs elsewhere.

• When an AI system has been demonstrably wrong in a consequential context and the organization has not developed a clear explanation for why, continuing to use it at the same level of trust is an operational error.

• When a model is updated or retrained by its provider, treat it as a new source and revalidate. Prior reliability does not transfer automatically.

• When you discover that the model has been systematically producing outputs shaped by the framing of your prompts rather than by evidence—that you have been leading the witness without realizing it—you may need to reset the relationship.

Burning a source is not a failure of the source-handling relationship. It is often the proof that the relationship was being handled well.

What This Means for How You Lead

The three of us came to this issue through intelligence work, but the problem is not limited to intelligence organizations. Any leadership environment where AI tools are proliferating faces the same structural challenge: the tools are fast, fluent, and confident, and organizational incentives often reward those who use them most rather than those who use them best.

The research bears this out. INSEAD’s 2025 analysis of firm-level AI adoption found that generative AI shifts value toward higher-order human judgment—not away from it. Microsoft’s research confirms that organizations with a well-calibrated understanding of AI perform better across missions than those that simply maximize usage. The tool is the easy part. The discipline is the hard part.

For leaders, the implications are practical:

• Build the habit of debriefing discipline before you scale AI adoption. The protocol above should be standard practice, not optional.

• Create psychological safety so people can flag AI errors. The greatest risk in any source-handling operation is the team member who saw the problem but didn’t say anything because the source had too much credibility.

• Distinguish between AI as a collection tool and as an analytical tool. Automate the former aggressively. Guard the latter carefully.

• Evaluate AI systems with the same rigor you would apply to any source—including periodic reviews of whether the relationship continues to produce reliable value.

Used with discipline, generative AI can be a genuinely powerful analytical partner—the kind of well-placed, high-access source that an experienced handler learns to work with carefully and derive real value from. Used without discipline, it becomes a certainty-destroyer: introducing noise, eroding judgment, and producing false confidence at scale.

The HUMINT model doesn’t make AI safer by limiting what it does. It makes AI safer by raising the standard for what we do with what it gives us.

AI doesn’t give you answers. It gives you reports. And reporting always requires a handler’s skeptical, trained eye.

About the Authors

The authors are former national security officers with combined experience across human intelligence operations, national laboratories, and management. Their views are their own and do not represent the position of the Central Intelligence Agency or the United States government.

Mike Mears is a leadership expert, bestselling author, creator of LeadCultureChange.com and former CIA Chief of Human Capital. As the founder of the CIA Leadership Academy, he trained managers and senior executives in practical leadership strategies grounded in neuroscience and human behavior. Mears holds an MBA from the Harvard Business School, and a BS degree from the US Military Academy at West Point.

James “Jim” Lawler served for 25 years as a CIA operations officer in various international posts and as Chief of the Counterproliferation Division’s Special Activities Unit. He was a member of the CIA’s Senior Intelligence Service from 1998 to 2005. Lawler was a specialist in the recruitment of foreign spies, and he spent over half of his CIA career battling the proliferation of weapons of mass destruction, including serving as the chief of the A. Q. Khan Nuclear Takedown team, which resulted in the disruption of the most dangerous nuclear weapons network in history.

John O’Neil, Ph.D. has extensive service in numerous leadership roles in academia and at Oak Ridge National Laboratory, where his work lay at the intersection of critical science and technology development, intelligence, issues of disruptive technical and WMD proliferation threats, and national security. He and his distinguished teams delivered numerous mission critical insights and solutions for intelligence, defense, and homeland security.

Share

Subscribe now

Background

These findings have led the OODA Network toward a central conclusion: the future of AI adoption will not be determined by which organizations deploy the fastest, but by which organizations learn how to govern increasingly powerful and partially unknowable systems. In this view, resilience - not impetuous acceleration - becomes the true competitive advantage.

This interview took place against the backdrop of several operating assumptions that have emerged from OODA Network discussions, OODAcon panels, member Slack dialogue, and follow-on OODA Loop research over the last two years. These conversations have produced a growing consensus that the principal challenges surrounding Agentic AI are not technological.

Rather, they are organizational, cultural, governance, and strategic in nature.

As models become more capable and agents become increasingly autonomous, enterprises are discovering that traditional controls, organizational structures, and risk management frameworks were not designed for probabilistic systems exhibiting emergent behaviors.

The result? An environment in which capability growth is compounding faster than institutional adaptation. Several themes have repeatedly surfaced across OODA research and member discussions:

1. AI capability is outpacing governance. Agentic systems are becoming increasingly capable while mechanisms for control, observability, evaluation, and behavioral assurance remain immature.

2. The primary bottlenecks are organizational, not technical. Most challenges stem from culture, incentives, leadership assumptions, and governance structures rather than shortcomings in the underlying technology.

3. Enterprise history is repeating itself. CISOs, GRC leaders, and IT teams are experiencing pressures reminiscent of earlier cloud, mobility, and cybersecurity adoption cycles-where executive urgency often exceeds operational readiness.

4. Acceleration without governance creates fragility. Pursuing AI adoption solely for speed or competitive signaling can introduce hidden dependencies, increase operational risk, and amplify the likelihood of cascading failures.

5. Risk has shifted from theoretical to operational. Data leakage, destructive actions, mission drift, and unintended agent behaviors are increasingly becoming normal characteristics of deployment rather than isolated anomalies.

6. The “big one” is viewed as a matter of not if, but when. A highly visible AI-driven incident will eventually become the seminal post-mortem and organizing principle for future enterprise AI governance.

7. Unknown behaviors are inherent features of probabilistic systems. The challenge is not eliminating emergence but learning how to govern it, observe it, and respond to it.

8. Scenario planning and resilience matter more than speed. Effective acceleration is not defined by “move fast and break things,” but by building the capacity to absorb failures, adapt under uncertainty, and demonstrate reasonable standards of care.

9. Resilience - not unwieldy acceleration - becomes the true competitive advantage. The future of AI adoption will not be determined by which organizations deploy the fastest, but by which organizations learn how to govern increasingly powerful and partially unknowable systems.

10. Trust becomes the enabler of value creation. Organizations that can establish deterministic controls, governance frameworks, and behavioral guardrails will ultimately be able to deploy AI more aggressively - and extract greater value from it.

The OODA Network Interview: Josh Devon, CEO, Sondera.ai

“I think of it like Jurassic Park: Life finds a way. The agents are going to find a way. And the more capable the agent is, the more likely you will end up with an unintended consequence - especially at scale. So I think that is the challenge that everyone is grappling with.”

Read more

Since 2024, the OODA Network has pursued a broad but unique research thread: materials science as the cross-sector, interdisciplinary driver of competitive advantage across:

• Energy

• Semiconductors

• Biotechnology

• Advanced manufacturing

• Space

• Defense

• Human-machine interfaces

• Critical minerals

• Clean industrial systems, and

• Frontier computing.

OODA Network discussions and our material science research archive suggest that materials science is a convergent, tangible strategic advantage. If AI is moving from language to laboratories, from prediction to design, and from digital transformation to physical transformation, then material science convergence across industry sectors is the diffusion metric to watch.

In 2026, the deeper measure of AI diffusion is whether AI becomes a discovery engine for the physical world.

High-level Summary

“Materials science is solving far more than isolated scientific problems-it is reshaping industries.” - Scott Nuzum

In 2023, OODA Loop contributor Scott Nuzum contributed an analysis on: Five Exciting Breakthroughs in Materials Science. That post captured something that is often obvious to scientists and engineers, but is still somewhat under-appreciated by business leaders: materials science is one of the hidden operating systems of modern technological advantage.

Since then our material science analysis has consistently framed materials science as foundational infrastructure for the future economy. Our analysis eventually synthesized into a single strategic thesis: nations and organizations that accelerate the discovery, development, manufacturing, and deployment of new materials will shape the next era of global competition.

Read more

“The nation’s cognitive infrastructure includes the mental capacities of our citizens and the decision-making ability of people, organizations, and our government. It also includes the information channels used to feed our decision-making capabilities and the education and training systems used to prepare people and organizations for critical thinking.” - Bob Gourley, “America’s Most Critical Infrastructure is also Our Most Neglected Infrastructure”, September 2019

Between 2019 and 2025 OODA has consistently framed one of our longest-running research themes: Cognitive Infrastructure. Over time, we have also engaged a growing body of research on cognitive warfare, neuroscience, narrative intelligence, and most recently: early signals on “cognitive sovereignty”. The strategic throughline? The most important critical infrastructure may reside in human cognition itself.

What has emerged from our initial framing, our network discussions, and analysis has gone far beyond traditional discussions of cybersecurity, digital transformation, or information operations. From a systems perspective, these frameworks have informed the OODA network about a new operational environment - shaped by the convergence of artificial intelligence, algorithmic amplification, information ecosystems, institutional trust, narrative competition, and machine-speed decision-making.

High-level Summary

Protecting cognitive infrastructure may ultimately become as important as protecting cyber infrastructure.

For decades, infrastructure discussions focused on roads, energy grids, telecommunications, and cyberspace. However, adversaries increasingly target perception, trust, and narrative rather than physical systems.

Read more

The central question: how are exponential organizations and high-performance individuals applying accelerating technologies to create compounding advantage and pull away from competitors? We’re building the agenda around that idea, and it’s shaping up well. Two very senior transformational government leaders are already confirmed for keynote conversations, and we’re putting together a program designed specifically for change agents who want real insight into the technology-enabled future, not just trend reports.

This post provides two links for registering for OODAcon.

Share

The first is for non-OODA network members. Share this one with anyone you believe would benefit from being in the room. It provides a 15% discount off ticket prices:

www.eventbrite.com/e/1987659902274/?discount=OODAcommunity

As an OODA Network Member use your 50% discount code:

Read more

A vital precursor to any enterprise AI value creation/capture and business model transformation effort within your organization (tactically induced by AI acceleration) is a baseline mapping out of the systemic risks and opportunities created by emergent autonomous behaviors and agentic AI operating system design challenges.

About the OODA Network and OODA Loop Research

“The OODA Almanac 2026 charts these shifts, urging leaders to observe the new anchor points, orient decisively, and act to thrive amid the turbulence. The OODA Network stands ready to guide you through 2026’s storms. We are a community of decision-makers who understand that the future belongs to those who observe it first, orient fastest, and act with conviction.” - Matt Devost, OODA Almanac 2026 | Re-anchoring in the Age of AI

The underlying premise is simple: in an environment where the rate of change is accelerating, the organizations that thrive will be those that learn faster, adapt earlier, and act with greater clarity.

Join the OODA Network to track these forces together, subscribe to OODA Loop, and join us at OODACON 2026.

The OODA Network is a community of executives, technologists, security professionals, investors, researchers, and decision-makers focused on navigating periods of rapid change. Its value lies in helping leaders identify emerging risks and opportunities early, test assumptions against a trusted peer network, and improve the speed and quality of strategic decision-making.

OODA Loop Research supports that community of practice by tracking weak signals, disruptive technologies, geopolitical shifts, security risks, and changing enterprise operating models. The goal is not simply to predict the future, but to help leaders recognize when legacy assumptions are becoming liabilities and when new strategic anchor points are emerging.

The OODA Almanac 2026 describes this challenge as a process of re-anchoring. In an era of AI acceleration, nonlinear innovation, geopolitical volatility, and institutional stress, organizations must move beyond static planning cycles and rigid operating models. They need the ability to observe changing conditions, orient around new realities, decide with incomplete information, and act before the window of opportunity closes.

This is where the OODA Network creates distinctive value. It functions as a trusted micro-network for leaders operating in an increasingly noisy and synthetic information environment. Through research, convenings, peer exchange, and ongoing analysis, the Network helps members separate meaningful signals from distraction, compare perspectives across sectors, and develop more resilient strategies.

OODA Loop Research is designed to make that orientation actionable. Its work translates complex developments into practical frameworks, decision-support insights, and recommendations for leaders responsible for security, resilience, innovation, and enterprise transformation.

Join the OODA Network to track these forces together. Or subscribe to OODA Loop.

The High-Level Summary

“Of course, we have agents, and, of course, they work side by side, and they’re part of our workforce, and they’re incorporated into everything. Kind of like: of course I used a GPS to get here this morning.” - Panelist Josh Ray, OODA Network Member and CEO of Blackwire Labs @ OODAcon 2025, From Automation to Autonomy

The organizations that achieve the greatest value from Agentic AI will not be those with the most advanced models, the largest data sets, or the biggest AI budgets. And the next generation of Enterprise AI value creation will focus less on measuring technology investments - and more on measuring intelligence investments.

For business leaders, this means that the ultimate purpose of an Enterprise AI Agentic Operating System is not automation; it is value creation - amidst an unprecedented acceleration of enterprise IT business model transformation - enabled by the tracking and orchestration of agent-based emergent behaviors and architectures.

Read more

Note: This post captures content from the monthly OODA Network meeting and is therefore only for full OODA Network Members and is not available at the subscriber tier.

About the OODA Network Monthly Meeting

OODA hosts a monthly video call to help members optimize opportunities and reduce risk, discussing items of common interest to our membership. These highly collaborative sessions are always an excellent way for our members to meet and interact with one another while discussing topics such as global risks, emerging technologies, cybersecurity, and current or future events that impact their organizations. We also use these sessions to help focus our research and understand member needs. To encourage open discussion, these sessions are conducted under Chatham House rules, where meeting participants are free to use the information shared in the meeting but are asked not to quote or identify other speakers directly. We also maintain privacy when preparing summaries of these sessions, as seen in the one that follows.

The High-Level Summary

“Every enterprise in the world was built on one assumption: cognitive work scales with human hours. You want more analysis? Hire more analysts. More code? More developers. Generative AI has broken that ratio forever. And here’s the part that nobody wants to talk about out loud. Most of our operating models that are built for the world that we lived in yesterday are now broken.“

- Dell Technologies Vice Chairman and COO

The May 2026 OODA Network meeting focused on the accelerating impact of artificial intelligence across enterprise operations. Network members framed the discussion around the idea that AI is no longer a future technology wave but an active restructuring force already altering enterprise operating models.

Share

Read more

These early AI adoption industry sectors between 2021 and 2025 foreshadowed the architecture of emerging scenarios:

• Healthcare exposed operational fragility.

• Cybersecurity exposed persistent adversarial acceleration.

• Finance exposed the emergence of AI-native infrastructure and programmable economic coordination.

Together, they provide a real-world map of how acceleration transforms institutions, infrastructure, governance, trust systems, and geopolitical competition. The most important lesson from these early adopter sectors is that AI adoption rewires the surrounding ecosystem. Industries become increasingly autonomous, interconnected, data-dependent, cyber-physical, and geopolitically contested.

Share

Read more

The cybersecurity pool party is back: 6 August, Las Vegas

Our annual Las Vegas Cybersecurity Pool Party returns on 6 August from 5pm to 9pm at the Mandalay Bay Daylight Beach Club. Grammy-nominated DJ Morgan Page will once again be setting the tone.

If you’ve been before, you know the format. If you haven’t: picture a few hundred cybersecurity and technology executives, senior practitioners, and some of the world’s sharpest cyber analysts, all gathered poolside in a setting that’s deliberately relaxed. No presentations. No panels. Just great music and an enjoyable time with with people who are actually doing work in the community.

Year after year, this one produces the kind of connections and candid exchanges that formal conferences can’t replicate. It’s become one of the highlights of the Vegas cyber week calendar for good reason.

Full details and registration here.

OODAcon 2026: Escape Velocity 4 November, Reston

OODAcon will be held on 4 November at the Hyatt Regency Reston, and this year’s theme is Escape Velocity.

The central question: how are exponential organizations and high-performance individuals applying accelerating technologies to create compounding advantage and pull away from competitors? We’re building the agenda around that idea, and it’s shaping up well. Two very senior transformational government leaders are already confirmed for keynote conversations, and we’re putting together a program designed specifically for change agents who want real insight into the technology-enabled future, not just trend reports.

The full agenda will be published as it comes together. But now is the time to block the date.

OODA Network members get 50% off ticket prices.

Learn more and register here.

Questions about either event? Interested in learning about sponsorship opportunities? Reach out to us directly. We’ll see you there.

Share

The financial sector’s AI acceleration story is ultimately about more than banking, markets, or digital payments. It is about the emergence of a new operational architecture for global economic coordination.

This OODA Network analysis synthesizes the major themes emerging across OODA Network discussions and OODA Loop’s financial sector AI and exponential technologies research from 2021 to 2025 - and positions the financial sector as one of the earliest large-scale operational testbeds for AI-native institutional transformation.

The financial sector is rapidly emerging as one of the most consequential AI acceleration environments in the global economy. Unlike many industries still experimenting with isolated deployments, financial institutions are integrating artificial intelligence directly into market operations, fraud detection, cybersecurity, compliance automation, treasury management, trading systems, customer interaction, digital identity, and increasingly, the architecture of money itself.

At the same time, the sector is becoming a convergence layer for multiple compounding accelerations: AI, blockchain infrastructure, stablecoins, post-quantum cryptography, cross-border data regulation, geopolitical fragmentation, digital identity systems, autonomous financial agents, and systemic cyber risk.

This convergence is transforming financial infrastructure into one of the primary strategic operating environments of the AI era. As a result, the financial sector provides one of the clearest early case studies of how AI diffusion reshapes institutions, infrastructure, governance, cyber risk, operational tempo, and geopolitical competition simultaneously.

Share

Executive Summary

AI adoption is no longer primarily a technology story. It is now an infrastructure story. And in the financial sector, that infrastructure is rapidly becoming global, autonomous, interconnected, and strategically contested. The sector’s aggressive adoption curve provides an early signal for how AI diffusion may unfold across other industries. The financial sector is not merely adopting AI. It is helping define the operational architecture of the AI economy.

Read more