Regulators from the EU, US, Japan, and Brazil unveiled a joint framework for AI oversight on Tuesday, marking the first major attempt at global rules. The "Seoul Accords" require companies to disclose training data sources, test high-risk models for bias, and add watermarking to AI-generated content.
The rules classify AI systems by risk: "unacceptable" uses like social scoring are banned, while tools for healthcare or hiring face strict audits. Firms must report major incidents within 48 hours and allow third-party safety evaluations before release.
Meta, Google, and OpenAI said they support harmonized rules but asked for clearer technical standards. Smaller startups warned compliance costs could stifle innovation. Enforcement starts July 2027, with fines up to 6% of global revenue for violations. China and India participated in talks but have not signed, citing sovereignty concerns. The framework will be reviewed yearly as the tech evolves. Digital rights groups called it "a baseline, not a finish line" for AI safety.