AI Governance's Hidden Power: Why the Most Impactful Work Happens Behind Closed Doors

The piece argues that the most consequential AI governance work happens out of public view. Examples include who actually shaped the UK AISI's agenda, changed the AI Act to include GPAI model provisions, convinced Rishi Sunak to take AI risks seriously, organized meetings between Yoshua Bengio and executive branch officials, and drafted ambitious questions for the G7's Hiroshima Process questionnaire. These insider efforts—by people within ministries, international secretariats, and multilateral committees—often determine the direction of regulation and commitments, such as the Seoul Frontier Safety Commitment or the structure of the EU AI Office's Code of Practice.

The author, who transitioned from technical AI safety to governance, observes a systematic bias in the AI safety community toward visible work: open letters, press statements, and intellectual production. This over-indexing on the “outsider game” may cause the community to miss large avenues for impact. Insider work is less glamorous, often requires different skills (bureaucratic navigation, relationship-building), and can feel alien to technically minded professionals. Yet as the author notes, even when public pressure opens doors, someone still has to walk through them—and that someone is often invisible.