Anthropic CEO Dario Amodei called for mandatory third-party testing of frontier AI models on Wednesday. He also wants to enable governments to prevent systems that fail safety audits.
The article, titled “Policy on Exponential AI,” arrived one day after Anthropic released Claude Fable 5. The company accompanied it with a legislative proposal on model testing and a job displacement framework.
Dario Amodei moves from transparency to binding rules
Anthropic spent 2025 supporting disclosure-based laws. The company supported SB 53 in California, the RAISE Act in New York, and SB 315 in Illinois. However, Amodei declared that transparency alone no longer matches risk.
It proposes a system modeled after the Federal Aviation Administration (FAA). Models that exceed the computing threshold will face mandatory third-party audits in four areas.
They cover cybersecurity, bioweapons, loss of control, and automated AI research.
“Frontier AI models, such as aircraft, should be required to undergo technical testing and scrutiny, and their launch should be banned or reversed as a threat to public safety if they do not meet high safety standards,” Amodei said. books In the article.
Follow us on XTo get the latest news as it happens
The plan goes beyond the White House’s executive order in June to request Regarding artificial intelligence, which Amodei welcomed as incremental progress.
It also wants prompt reporting of safety incidents and strict protection of model weights.
Cyber risks put cryptocurrency infrastructure under scrutiny
Amodei described cybersecurity as the first risk to be fully realized. He pointed to the Cloud Mythos preview, which solved 73% of Expert-level cyber challenges It has never passed any artificial intelligence before.
The article warns that frontier models could disrupt the financial sector and critical infrastructure.
BeInCrypto analysis is also tagged Security risks of DeFi Related to the Mythos class models, since decentralized finance (DeFi) protocols hold open value and are vulnerable to attack.
Meanwhile, Anthropic shipped Claude Fable Model 5 on June 9 with safeguards preventing high-risk cyber and biological requests.
Amodei says such voluntary limits cannot replace binding rules across the industry.
He also warned that risks of autonomy could follow. Anthropic’s own data already appears AI builds better AIwhere Claude wrote most of the code in major artificial intelligence laboratories.
Loss of jobs, civil liberties and the Democratic Alliance
On the economic front, the article suggests wage insurance, tax incentives for job retention, and workforce training grants.
If displacement proves persistent, Amodei says UBI could be funded through corporate taxes or capital gains.
The civil liberties agenda is equally mentioned. Amodei wants autonomous weapons to be completely banned by local law enforcement. It also urges Congress to close the data broker loophole that enables bulk surveillance purchases.
Geopolitically, it calls for the creation of a coalition of democracies to control chips and semiconductor manufacturing equipment.
He cites the pending US bills MATCH and OVERWATCH as first steps toward more stringent and coordinated export controls.
Amodei dismissed the idea that public fear of artificial intelligence is a marketing problem, describing this concern as accurate. Congressional acceptance of Anthropic’s test proposal may now determine the next phase of AI policy.
this post Dario Amodei demands power to ban unsafe AI a day after Claude’s Fable 5 release appeared first on BeInCrypto.
