After two weeks of intensive negotiations in Geneva, representatives from 47 nations signed a framework that establishes mandatory safety testing, transparency requirements, and cross-border incident reporting for advanced AI systems.
What the treaty covers
The agreement creates a new international body — the AI Safety Observatory — tasked with coordinating audits of models above a defined compute threshold. Companies must submit red-team reports before deploying systems in participating countries.
"This is the nuclear non-proliferation treaty of our era," said the summit chair. "We are not banning innovation — we are ensuring it serves humanity."
India, the European Union, and Japan were among the first signatories. Industry groups welcomed clarity but warned that compliance costs could slow startup innovation in emerging markets.