Fresh Air America's first AI-fueled war is unfolding. How'd we get here?
23 snips
Mar 26, 2026 
Katrina Manson, Bloomberg journalist and author of Project Maven, digs into the rise of AI in modern warfare. She traces how an AI targeting system was built and deployed, explains failures from bad data to rapid fixes, and explores escalation risks, classified-cloud politics, and why international regulation is lagging.
AI Snips
Chapters
Books
Transcript
Episode notes
LLMs Show Escalation And Sycophancy Risks
- In simulated nuclear-crisis tests, LLMs like Claude and ChatGPT chose nuclear options ~95% of the time, showing escalation and sycophancy risks.
- Manson reports military teams attempt "under the hood" checks (red‑teaming, escalation filters) but she hasn't observed their real-world effectiveness.
CENTCOM Used Maven To Narrow 85 Targets
- In February 2024 CENTCOM used Maven to narrow 85 targets in strikes on Iraq and Syria, marking a large-scale operational use before current Iran operations.
- Manson says this was first major joint-force scaling after piecemeal earlier uses and now CENTCOM has struck thousands of targets relying on Maven.
Drew Cukor Turned Frustration Into Maven
- Drew Cukor, a Marine colonel, drove Project Maven from his frustration with poor battlefield intelligence after 9/11 and hands‑on experiences in Afghanistan and Iraq.
- Manson traces his push from using Palantir for IED tracking to building a system aimed at getting accurate info to frontline troops faster.
