Risky Bulletin Between Two Nerds: How NSA will use AI
Feb 23, 2026
A deep look at how elite intelligence services will adopt AI cautiously, favoring stealth and correctness over reckless use. Discussions cover AI-powered espionage trends, task decomposition for intelligence workflows, and which cyber roles are most AI-friendly. They explore AI for continuous monitoring, mapping target networks, speeding coding and malware development, and how AI might complicate attribution.
AI Snips
Chapters
Transcript
Episode notes
Why Five Eyes Won't 'YOLO' AI
- Professional espionage agencies value correctness and stealth over risky rapid experimentation with AI.
- Tom Uren argues NSA/ASD will avoid 'yolo' AI use because mistakes risk exposure and they place a premium on correctness.
Small Tasks Let AI Increase Security Reliability
- Breaking big tasks into small verifiable steps is how AI adds robust value to security workflows.
- The Grugq cites Trail of Bits using AI to generate test cases and micro-tasks that are easy to check and low risk.
Use AI To Double Check Risk Assessments
- Use AI as a second-eye for assessments rather than the sole author of strategic decisions.
- The Grugq suggests running parallel AI risk assessments then compare-and-contrast to catch missed angles for free.