Anthropic CEO Dario Amodei called for mandatory third-party testing of frontier AI models on Wednesday. He also wants governments empowered to block systems that fail safety audits.
The essay, Policy on the AI Exponential, arrived one day after Anthropic released Claude Fable 5. The company paired it with a legislative proposal on model testing and a job displacement framework.
Dario Amodei Moves From Transparency to Binding Rules
Anthropic spent 2025 backing disclosure-based laws. The company supported SB 53 in California, the RAISE Act in New York, and Illinois’ SB 315. However, Amodei announced that transparency alone no longer matches the risks.
He proposes a regime modeled on the Federal Aviation Administration (FAA). Models above a compute threshold would face mandatory third-party audits in four areas.
These cover cybersecurity, biological weapons, loss of control, and automated AI research.
“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,” Amodei wrote in the essay.
Follow us on X to get the latest news as it happens
The plan goes beyond the White House’s June Executive Order on AI, which Amodei welcomed as incremental progress.
He also wants prompt safety incident reporting and strict protection of model weights.
Cyber Risks Put Crypto Infrastructure on Notice
Amodei called cybersecurity the first risk to fully materialize. He pointed to Claude Mythos Preview, which solved 73% of expert-level cyber challenges that no AI had passed before.
The essay warns that frontier models could disrupt the financial sector and critical infrastructure.
BeInCrypto analysis has likewise flagged DeFi security risks tied to Mythos-class models, since Decentralized Finance (DeFi) protocols hold open, attackable value.
Meanwhile, Anthropic shipped the Claude Fable 5 model on June 9 with safeguards that block high-risk cyber and biology requests.
Amodei argues such voluntary limits cannot substitute for binding rules across the industry.
He also warned that autonomy risks may follow. Anthropic’s own data already shows AI building better AI, with Claude writing most of the code at major AI labs.
Job Losses, Civil Liberties, and a Democratic Coalition
On the economic front, the essay proposes wage insurance, retention tax incentives, and workforce training grants.
If displacement proves enduring, Amodei says universal basic income could be financed through company or capital gains taxes.
The civil liberties agenda is equally pointed. Amodei wants fully autonomous weapons banned from domestic law enforcement. He also urges Congress to close the data broker loophole that enables bulk surveillance purchases.
Geopolitically, he calls for a coalition of democracies to control chips and semiconductor manufacturing equipment.
He cites pending US bills MATCH and OVERWATCH as first steps toward tighter, coordinated export controls.
Amodei rejected the idea that public fear of AI is a marketing problem, calling the concern accurate. Whether Congress takes up Anthropic’s testing proposal may now define the next phase of AI policy.