Anthropic CEO backs mandatory AI testing with $350 million pledge

Dario Amodei wants governments to block dangerous AI models before they reach the public — and he's putting $350 million behind the push.

Dario Amodei, chief executive officer of Anthropic, on June 11 proposed mandatory third-party testing for frontier AI systems exceeding 10^25 FLOPs of compute or $500 million in AI revenue, with government authority to block deployment if models pose catastrophic risks across four categories.

"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 said in a blog post titled "Policy on the AI Exponential."

The framework targets cybersecurity, biological weapons, loss of control and automated research and development risks. Anthropic committed $200 million for policy research and $150 million for scholarships to support the initiative, which applies to any company meeting the compute or revenue thresholds — a group that includes OpenAI, Google DeepMind and Meta Platforms.

The proposal marks the most aggressive regulatory framework backed by a major AI CEO and places Anthropic in direct contrast with the Trump administration's voluntary testing approach. President Donald Trump signed an executive order June 2 that gave the intelligence community an enhanced role in model testing but stopped short of requiring pre-approval before deployment.

Amodei's call for binding regulation comes as his own company's Claude Mythos Preview model demonstrated the accelerating threat landscape. Anthropic said Mythos built 16 working exploits targeting Firefox and Windows within hours, including eight full privilege-escalation exploits for Windows kernel vulnerabilities in 18 hours. The cost of creating those exploits was $15,700 in API credits, or roughly $2,000 per chain.

"The binding constraint to N-days is now just a few thousand dollars and API access, which expands the pool of capable N-day attackers dramatically," Anthropic said in a separate report. The company urged organizations to shift from "N-day" to "N-hour" patching timelines, warning that industrial control systems, medical devices and internet-of-things devices face heightened exposure as exploit costs approach zero.

Beyond safety testing, Amodei proposed economic measures to address potential AI-driven labor displacement, including universal capital accounts, wage insurance and employer retention incentives. Sen. Bernie Sanders last week proposed a 50 percent tax on stock offerings of top AI companies that would give the government a voting stake, while Trump days earlier floated an unspecified profit-sharing partnership with leading AI firms.

Amodei also called for reforming regulatory institutions to accelerate AI-enabled innovation, suggesting the Food and Drug Administration could approve AI-discovered medicines faster through simulated trials and synthetic control groups. He backed continued international cooperation among democratic nations on AI governance and advocated a ban on domestic deployment of fully autonomous weapons systems.

"We now, globally and collectively, need to activate a slow and rickety policy apparatus to deal with risks and opportunities that are going to compound surprisingly quickly from here," Amodei wrote.

The current average U.S. tariff on Chinese goods stands at roughly 19 percent after the May 2025 escalation, according to the Peterson Institute for International Economics. The previous 10 percent increase in September 2024 reduced bilateral trade by $63 billion over six months, Census Bureau data show — a precedent for how regulatory friction can reshape cross-border technology flows if AI model testing requirements create similar barriers.

This article is for informational purposes only and does not constitute investment advice.