Back
Anthropic CEO Dario Amodei urges power to block risky frontier AI models

Anthropic CEO Dario Amodei urges power to block risky frontier AI models

Amodei says AI mastering the final 10% of tasks could enable full job automation, while a new report highlights the scale of revenue needed to sustain frontier AI investment.

Fact Check
The primary source — Amodei's own essay 'Policy on the AI Exponential' — directly states the government should have authority to block deployment of unsafe models, confirming the claim. This is corroborated by Axios and CryptoBriefing, which report the same core proposal of binding government authority to block or reverse risky AI model releases following mandatory safety testing.
    Reference: 123
Summary

Anthropic CEO Dario Amodei has called for governments to have the authority to block risky frontier AI models and backed mandatory testing for advanced systems, marking one of the strongest regulatory stances from a major AI lab. He has argued that frontier models pose "very real risks" to the financial sector, critical infrastructure and national security, and wrote that "Mythos Preview scrambled the global cybersecurity landscape." In a Bloomberg interview, he also warned that if AI masters the final 10% of tasks within jobs, it could trigger full job automation, raising the prospect of broad economic and social disruption. A new report from Crypto Briefing adds that Anthropic's revenue ambitions underscore the precarious economics of frontier AI development, saying Dario Amodei warned the company would need $1 trillion in revenue to survive, a claim that points to the immense capital demands behind leading AI models and the potential knock-on effects for tech sectors and investor confidence.

Terms & Concepts
  • frontier AI models: The most advanced AI systems at the leading edge of current development.
  • mandatory testing: A regulatory requirement that advanced AI systems undergo formal evaluation before release or wider deployment.
  • tiered AI release approach: A deployment model that gives different levels of access or safeguards to AI systems based on their capabilities and risks.