When Should AI Treaties End? Paper Reveals Practical Dilemmas of Treaty Termination

DavidSKrueger · x · 2026-07-22

When discussing halting AI development, the question "Until when?" frequently arises. AI safety researcher David Krueger highlighted a new short paper by Lennart Finke, exploring how to design termination conditions for an international treaty limiting AI development.

Krueger noted this is the first paper he's aware of that actually examines how this works in practice. The findings are intuitive: most treaties lack the detailed termination conditions often demanded of AI agreements. Treaties largely formalize agreements, provide monitoring mechanisms, and specify enforcement terms. However, signatories can and do violate or withdraw based on their perceived interests. Ultimately, this interest-based perception is what causes a treaty regime to persist or collapse, regardless of its written terms.

Original post →

More from AGI Musings

AGI Musings channel →