Sources & Methods Inside a secret Pentagon effort to bring AI to the battlefield
85 snips
Mar 23, 2026 
Katrina Manson, Bloomberg reporter and author of Project Maven, explores the Pentagon’s secret push to field AI in warfare. She recounts Project Maven’s origins, early technical stumbles and life-saving successes. She examines culture clashes inside the military, the ethics fights including tech worker protests, and how AI is reshaping targeting and escalation risks.
AI Snips
Chapters
Books
Transcript
Episode notes
Funny And Dangerous Failures Of Early Models
- Early computer vision models for Maven were trained on civilian images and misclassified battlefield objects (clouds as buses, trees as people).
- Analysts often rejected the tool until it produced clear wins like spotting hidden people faster than humans.
How Small Wins Created Momentum For Maven
- Successes (AI spotting a farmer or Marines faster) built trust and justified further field testing called 'field to learn.'
- Cukor recruited reservists who acted like an 'insurgency' inside the Pentagon to push adoption.
Using Human Networks To Bypass Pentagon Bureaucracy
- Bureaucratic resistance forced Maven's team to rely on personal networks and veteran relationships to pilot their tech.
- The group used Marine culture and reserve call-ups to work aggressively despite being small and underpowered.
