A Red Line and Oversight Framework for Government AI Contracts

LessWrong recently published a governance framework from TurnTrout, an AI researcher who left Google DeepMind, addressing how AI companies should handle government contracts involving targeting or surveillance. The framework proposes two red lines: AI cannot power autonomous targeting or force without meaningful human control evaluated case-by-case, and AI cannot convert bulk data into individualized intelligence on people who aren't already specific investigation targets. To enforce these standards, he proposes a Defense AI Review Body of seven senior staff members who audit contracts, issue findings, and produce annual transparency reports to employees with a tally of leadership overrides on flagged non-compliance. The framework is intentionally temporary, designed to be superseded if Congress passes legislation on autonomous weapons or AI surveillance.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton