On June 10, 2026, Anthropic published “Policy on the AI Exponential,” laying out its Advanced AI Framework, a proposal for how governments should regulate the most capable AI models. The framework’s central ask is that governments be given the legal authority to block or deter the deployment of models that pose a significant risk of catastrophic harm, such as enabling cyberattacks or the development of biological weapons, paired with civil penalties scaled to a company’s revenue.
The proposed rules are deliberately narrow in scope. They would apply only to models trained using more than 10 to the 25th floating-point operations, and only to developers with substantial AI activity, on the order of 500 million dollars or more in AI revenue or 1 billion dollars or more in AI research and development spending. In practice that captures a handful of frontier labs. Covered developers would conduct safety testing, publish a summary of results, file regular risk reports, and engage at least one qualified independent evaluator.
The framework is a policy proposal, not law. It is notable because it comes from a leading AI lab voluntarily asking for binding government authority over its own products, a contrast with the lighter-touch, largely voluntary approach the US executive branch set out earlier in the month. For business leaders, it signals where serious frontier developers think mandatory regulation is heading: compute and revenue thresholds, independent evaluation, and government power to halt deployment.