Anthropic CEO Dario Amodei has published an essay calling for stricter AI regulation in the United States and a shift from simple transparency to mandatory safety tests. Alongside the essay, the company presented two documents: an "Advanced AI Framework" for overseeing frontier models, and an "Economic Policy Framework" to prepare the economy for workforce displacement. Amodei proposed mandatory external testing for models that exceed a threshold of 10²⁵ FLOPs, or if a developer earns more than $500 million per year from AI or spends over $1 billion on AI research and development. The tests should cover cybersecurity, the creation of biological weapons, loss of control, and automated R&D. If unacceptable risks are identified, authorities should have the power to block the model's deployment — a measure Amodei compared to technical audits in aviation.

On the economic front, Amodei warned of potentially massive labor market disruptions, with unemployment possibly reaching 10% or higher. As countermeasures, he proposed improved data collection, retraining programs, and, as a last resort, the introduction of unconditional basic income. To support research in this area, Anthropic will allocate $200 million to an Economic Futures Research Fund and create a $150 million fellowship program for specialists.

The essay also pays special attention to civil liberties. Amodei called for establishing accountability rules for fully autonomous weapons, banning their use within the United States, and closing loopholes regarding data purchases for mass surveillance. He also proposed ensuring that citizens have access to AI no less capable than what the state uses for regulatory purposes. These initiatives come in the wake of the recent release of the Claude Fable 5 and Mythos 5 models, as well as statements from the Anthropic team about signs that AI is approaching recursive self-improvement, which may require mechanisms to slow down development.