The AI Policy Podcast
Episode: Inside Project Maven and AI-Powered Warfare with Katrina Manson
Date: March 26, 2026
Host: Gregory C. Allen, CSIS Wadhwani Center
Guest: Katrina Manson, Bloomberg journalist and author of "Project: A Marine Colonel, His Team and the Dawn of AI Warfare"
Episode Overview
In this episode, Gregory C. Allen sits down with journalist and author Katrina Manson to discuss her new book on Project Maven, a pivotal initiative in the U.S. military’s adoption of artificial intelligence (AI) for warfare. Their conversation explores Maven’s origins, technological triumphs and setbacks, ethical debates, operational transformations, and the broader implications for U.S. defense and geopolitics. Manson and Allen provide a behind-the-scenes look at how passion, bureaucracy, and rapid innovation converged to shape the “Microsoft Windows of warfighting” and how its influence reaches today’s conflicts, including the wars in Ukraine and Iran.
Guest Background: Who Is Katrina Manson? (01:21–02:28)
- Katrina Manson is a journalist, currently at Bloomberg covering AI, national security, and cybersecurity.
- Formerly with the Financial Times for 11 years, Pentagon correspondent (2017–2020), and covered intelligence and foreign policy.
- Prior experience as East Africa correspondent and reporting from conflict zones in Africa.
- Interest in military AI grew from frustration over lack of transparency and the ethical debate swirling around Pentagon AI initiatives, particularly Project Maven.
“Anyone who is animated by passions, as I discovered everyone to do with Project Maven is animated by passion, whether they're for it or against it, really.”—Katrina Manson [04:05]
Why Project Maven? The Need for AI in Modern Warfare (02:42–06:04)
- Manson was intrigued by Pentagon secrecy and protests by Google employees.
- Sought to unravel the debate about the implications of military AI, operator protection, civilian impact, and the ethical boundaries; transparency was a core motivation.
- Wanted to understand not only motivations but the practical, operational effects of deploying AI.
The Impact: Where Maven Stands Today (06:04–07:48)
Manson reads from her book about Maven’s extraordinary reach, illustrating its transformation from a niche project to a core element of U.S. and NATO military operations:
"Maven Smart System MSS...is now deployed in every branch of the US military and all over the world, incorporating more than 150 data feeds and the work of more than 50 companies. NATO started using a version of the system in the spring of 2025...Maven has already sped up the pace of war...went from being able to hit under 100 targets to...5,000 targets a day."—Katrina Manson [06:04]
Key Points:
- MSS is the software backbone for target development in multiple domains—including submarines, space, and drone operations.
- Adopted internationally; up to 10 NATO members lining up to adopt.
- AI now increases strike capabilities fivefold with help from large language models (LLMs).
- Used in high-stakes areas, including systems intended for defense of Taiwan.
Origins: Drew Cukor and the Birth of Maven (07:48–12:49)
- Drew Cukor: Marine Colonel, central figure and initial chief of Project Maven (till late 2021/early 2022), described as intense and passionate.
- Cukor’s frustration with intelligence-sharing shortcomings during Afghanistan deployments post-9/11 set the stage.
- Early efforts: wanted to log IED (improvised explosive device) attacks with real data, partnering with Palantir in 2011 to deliver better software to Marines.
- Contrasted old, ineffective analytic tools (Word, Excel) with forward-deployed, iteratively improved software—shaping his vision for AI.
“Many, many people talk about getting C cored and to Cukor and some points, you know, if you are working hard, sleeping not very much, and being very exacting...that certainly was one way he was described.”—Katrina Manson [08:55]
The Data Deluge and the Need for AI (12:49–16:56)
- Shift from lacking intelligence to having overwhelming amounts of drone imagery and video, far beyond human analytic capacity.
- Operators were missing actionable insights because data from drones went unexamined.
- Project Maven began as an attempt to automate the identification of objects in drone footage, initially attempting simple, labor-saving tasks (counting cars/people).
- Will Roper (DoD official) ran parallel AI projects on satellite imagery; Maven drew inspiration and funding from these efforts.
“The drones...are collecting information that is not actually being looked at. They're running so many videos...no one was actually then observing, analyzing, taking this data from the drone feeds.”—Katrina Manson [13:35]
Enlisting Silicon Valley: Ambition, Skepticism, and Google’s Role (16:56–22:36)
- Cukor’s vision always stretched far beyond humble beginnings—he “had this thesis...for something that could automate or could bring intelligence to the front lines.” [16:56]
- Prioritized embedding top AI talent and technology; actively recruited Google (wanted DeepMind/Google Brain, got Google Cloud), Palantir, Amazon, and startup Clarifai (led by 2012 ImageNet competition winner Matt Zeiler).
- Zeiler’s journey: from developing bridal-dress-identifying algorithms to working with DoD—motivated by claims that AI could save lives.
“He laid out the scenario where US Military operators...were concerned that they were getting attacked...turned out it was by cattle. If they just had AI that could be analyzing and observing...AI would help.”—Katrina Manson [21:10]
The Google Exit and AI Ethics Backlash (22:36–26:56)
- Google’s withdrawal after internal protests was a shock and a catalyst—brought ethics to the forefront and shifted public debate to U.S.–China AI rivalry.
- Pentagon “Jake” (Joint AI Center) leaned hard on responsible AI and outreach (“listening tour/apology tour” led by Jack Shanahan); sought to reassure the tech sector and public about non-lethal uses for AI.
- Maven and Jake operated in parallel—Maven on operational/technical delivery, Jake on public policy and “AI Ethics” infrastructure.
“Really, this is Shanahan reaching out to try and temper the water on AI, deliver the public acceptability of AI in warfare...Jake focuses very much on ethics...which annoys some people in the Pentagon.”—Katrina Manson [25:10]
Field “Deem to Learn”: Early Deployments and Setbacks (27:55–31:32)
- Operators’ acceptance was mixed: early Maven deployments, particularly in Somalia, often met with disappointment—glitchy, unreliable object-detection, confusing interfaces, resistance to change.
- Feedback loop: Maven relied on sending engineers to the field to coach users and rapidly update software, but institutional inertia, data formatting, and inter-system rivalry hampered integration.
- Cukor downplayed algorithmic perfection in favor of building a robust, operational system for future improvements.
“[Maven AI] was identifying too many things. The boxes were flashing up a lot...operators turned it off.”—Katrina Manson [28:50]
"Kukor...described the AI as a bag of potato chips...meant, of course the algorithms are no good. We don't care about the algorithms so much as the system..."—Katrina Manson [30:20]
Ukraine Invasion: Maven's Coming of Age (31:32–39:49)
- By Russia’s 2022 invasion of Ukraine, Maven was in bureaucratic flux (moving between agencies), but was rapidly pulled into action.
- Initially used for situational awareness (e.g., counting refugees), then shifted to identifying Russian targets for Ukrainian forces.
- Training data issues: algorithms trained on Middle East operations struggled in snowy Ukraine; rapid feedback and overnight retraining enabled quick adaptation.
- Developed key trust relationships with Ukrainian operators—sometimes leading to intelligence sharing so rapid that a Russian missile launcher was destroyed 18 minutes after AI detection [41:54].
- U.S. never directly told Ukrainians what to strike, but highlighted “points of interest” that became highly actionable.
“Ukrainians, in one case I was told about, couldn’t tell what they would be hitting from their own intelligence sources. And the Americans could say, trust us, hit it.”—Katrina Manson [36:26]
Integration Challenges: Technology, Bureaucracy, and Bandwidth (43:55–45:33)
- Success in Ukraine highlighted underlying tech challenges: network bottlenecks, data packet loss, and encryption slowdowns in U.S./NATO infrastructure were all major headaches.
- Maven’s demand for robust infrastructure accelerated broader modernization but revealed ongoing dependencies and vulnerabilities.
"packets of data were crisscrossing the Atlantic twice or even four times. And so they could lose packets of data...In order to have the classified systems running, you need to use encryptors...That created a bottleneck."—Manson [44:41]
Magic of Maven: Is It the AI or the System? (42:44–47:40)
- Debate within Maven: was the real breakthrough the underlying AI, or was it the data integration and easy-to-use interface (Palantir’s role)?
- Some insiders argue the platform could swap out AI models or interface providers but the fusion of data and speed is what made it transformative.
- Still faces competition (e.g., with NRO) and operational challenges on the “last mile” (user/operator level).
The Iran War: Modern Day Use & LLMs (48:04–50:40)
- Maven Smart System central to current operations, integrating LLMs like Claude from Anthropic to massively accelerate data-processing workflows.
- LLMs’ role is in streamlining the “admin side” of the targeting cycle—handling approvals and paperwork, not decision-to-strike.
- CENTCOM has publicly credited Maven and AI with reducing targeting cycles from days/hours to just seconds in some cases.
"He’s taking time to say that in the middle of this war. So their focus, their belief in leaning forward into AI is fascinating."—Manson [49:23]
Lessons, Legacy, and Ongoing Debates (50:40–59:59)
- Maven is unique for its rapid operational focus, scale, and blending of public-private partnership.
- Ongoing debate: should Maven focus on a single platform, or push the bleeding edge of embedded AI? Some team members (e.g., Colin Carroll) felt user interfaces distracted from core AI research.
- Persistent organizational rivalries (with NRO, services), vendor lock-in dilemmas, and challenges of AI on autonomous platforms (e.g., sea drone vision issues).
- Maven has succeeded in making itself indispensable and expanding globally, but questions remain about long-term integration and operator-level effectiveness.
Case Study: Admiral War Whitworth’s Transformation (57:01–59:59)
- Whitworth, initially a fierce Maven skeptic, was converted by the system’s remarkable adaptability in war—updating and responding faster than any previous system.
- Championed reliability metrics for AI, expanding deployment especially in the Indo-Pacific theater (INDOPACOM).
- Under his leadership, adoption spread, annual AI summits launched, and senior commanders became enthusiastic champions.
"He found that Maven was able to update and respond to the realities of war quicker than anything he'd ever seen. And it was that pliability of the software..."—Manson [57:49]
Notable Quotes & Memorable Moments
- “[Maven is like] the Microsoft Windows of warfighting…”—Gregory Allen [05:51]
- “Anyone who heard that and doesn’t understand why this is a vitally important story to tell is clearly clueless.” —Gregory Allen [07:48]
- “To me, it isn't as clear a picture as the one you're presenting...I've also heard that Maven is not succeeding at the operator level, which is exactly the level it was hoped for because of bandwidth problems.” —Katrina Manson [54:09]
Timestamps for Major Segments
- 01:21 — Katrina Manson’s background and entry into military AI journalism
- 06:04 — Manson’s book excerpt: Maven’s global reach and operational impact
- 07:48 — Origins: Cukor’s Afghanistan trauma and software revolution
- 13:35 — Data deluge and “actionable intelligence” bottlenecks
- 16:56 — Commercial AI partnerships and Google protests
- 22:36 — Google’s exit, ethics backlash, and the rise of “responsible AI”
- 27:55 — Field experiments, operator resistance, and rapid iteration
- 31:32 — Ukraine: Maven’s first real test in peer conflict
- 43:55 — Infrastructure and integration headaches
- 48:04 — Iran war and LLM-enabled targeting cycles
- 57:01 — Admiral Whitworth’s journey from skeptic to Maven advocate
Conclusion
Katrina Manson’s reporting on Project Maven reveals a complex, passion-driven saga—the story of a military experiment that has grown into a critical backbone for modern warfare, transforming how intelligence and lethal action interface with data and automation. The path was rocky, marked by skepticism, operational setbacks, and ethical controversies, but the result has been a platform now embedded across American and allied militaries. As AI advances and geopolitics heats up, Maven is both a bellwether and a battleground for debates about speed, transparency, risk, and the future of conflict.
[For those interested in the gritty details, nuanced personalities, and internal conflicts, Katrina Manson’s book comes highly recommended by Gregory Allen and is positioned as essential reading for anyone serious about military AI.]
