AI Constitution
1

Thesis

Non-Proliferation of Dangerous AI

Sign in to propose an editHistory

A global non-proliferation treaty should prevent the development of dangerous AI.

Summary

The case for including this

A global non-proliferation regime treats the most dangerous AI like other catastrophic technologies, recognizing that unilateral restraint fails without coordinated, verifiable international limits. By naming a treaty, the principle points toward concrete governance machinery rather than vague aspiration, and it acknowledges that race dynamics make voluntary safety unstable. For risks that cross borders, only collective constraint can be credible.

The case for changing or excluding this

Including this trades national sovereignty and competitiveness against collective safety, and an unenforceable treaty disarms the cautious while emboldening defectors who race ahead. The clause also states a geopolitical policy prescription rather than a principle for how AI should behave; 'dangerous AI' is undefined; and non-proliferation analogies to nuclear or bio weapons may not fit a software technology that is cheap to copy and dual-use. It belongs in policy advocacy unless reframed as a principle about AI itself, tempered by realism about whether such a regime can be verified.

Discussion

Sign in to join the discussion.

Related resources