Oct 14, 2026 · 34 min
AI safety advocate says competition defeats voluntary restraint
AI Safety: ‘Criminals don’t self regulate’ | Connor Leahy
The episode examines whether governments can slow the race toward potentially uncontrollable AI before regulation arrives too late.
- 1Connor Leahy traces his shift from open-source model building to campaigning for restrictions on superintelligence.
- 2Voluntary pauses cannot constrain competitors, making enforceable rules and national-security oversight central to AI safety.
- 3Political pressure in the United States and United Kingdom could determine whether superintelligence is prevented or regulated after development.
Don't miss
Leahy distinguishes between a future that takes extinction risk seriously and one that allows unrestricted AI development, arguing that political action determines the outcome.
The brief
Connor Leahy moved from co-founding EleutherAI and leading Conjecture to advocating restrictions on superintelligence, after changing his view of AI’s trajectory and risks.
His core argument is that companies competing for power cannot reliably police themselves: one firm’s pause leaves rivals free to continue, while harms may already have occurred.
Leahy describes taking the fight to Washington, D.C., where briefing congressional offices and national-security officials has shown him that politicians can be persuaded.
The proposed response is preventive rather than reactive: define superintelligence by capability, criminalize its development, and build bipartisan support before such systems exist.
The episode’s sharpest turn comes when Leahy contrasts a world that treats extinction risk seriously with one that permits an unchecked race for increasingly capable AI.
For Leahy, parliamentary debate, public pressure, and contact with elected representatives are not symbolic extras but the mechanisms that could create enforceable safeguards.