Anthropic CEO Urges Mandatory Audits, Govt Power to Block AI

Anthropic CEO Dario Amodei published an essay on June 10, 2026, a day after the company released Claude Fable 5. The essay sets out regulatory and policy changes the company wants for so‑called frontier AI models.

Amodei proposed a regulatory regime modeled on the Federal Aviation Administration. Under the plan, models that exceed a set compute threshold would face mandatory third‑party technical testing. Regulators would be able to block or reverse a model’s release if it fails safety audits.

The essay identifies four audit areas for frontier systems: cybersecurity, biological threats, loss of control, and automated AI research. It also calls for prompt reporting of safety incidents and strict protection of model weights to limit unauthorized re‑use or exploitation.

Amodei argued that AI capabilities are advancing faster than current policy can handle, writing, “AI is progressing extremely fast-much faster than the policy process was built to handle.” He added that “Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.”

The essay cites cybersecurity as the first risk already showing up. Anthropic said an earlier internal model, Claude Mythos Preview, solved about 73% of expert‑level cyber challenges that previous systems had not solved. The paper warns that similar capabilities could threaten financial services, critical infrastructure, and decentralized finance protocols that expose value in publicly accessible code. Anthropic said Claude Fable 5 includes safeguards to block high‑risk cyber and biological requests.

On economic policy, the essay proposes wage insurance, tax incentives for employers to retain workers, and grants for retraining. It notes that if displacement proves long lasting, universal basic income could be financed by taxes on companies or capital gains.

On civil liberties, the essay calls for a ban on fully autonomous weapons for domestic law enforcement and for closing a data broker loophole that enables bulk surveillance purchases. Internationally, it urges a coalition of democracies to restrict exports of chips and semiconductor manufacturing equipment and cites pending U.S. bills MATCH and OVERWATCH as initial steps.

The essay welcomed a recent White House executive order on AI as a limited advance and urged that enforceable testing rules be added to the policy mix. Anthropic paired the Claude Fable 5 rollout with a legislative proposal on model testing and a framework addressing job displacement.