Anthropic CEO Calls for Binding AI Safety Rules, Cites Cybersecurity as First Materialized Risk

Anthropic CEO Dario Amodei published an essay Wednesday calling for mandatory third-party testing of frontier AI models and government authority to block systems that fail safety audits. The essay is titled “Policy on the AI Exponential.”

Amodei Proposes FAA-Style Mandatory Audits Across Four Risk Categories

Amodei’s proposal would require models above a compute threshold to undergo mandatory third-party audits covering four areas: cybersecurity, biological weapons, loss of control, and automated AI research. 

The framework is modeled on Federal Aviation Administration oversight. “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. 

The proposal goes beyond the White House’s ‘Promoting Advanced Artificial Intelligence Innovation and Security’ Executive Order on AI, signed June 2, 2026, which Amodei described as incremental progress. He also called for prompt safety incident reporting and strict protection of model weights. Anthropic published the essay alongside a legislative proposal on model testing and a separate job displacement document. 

In 2025, Anthropic backed disclosure-based laws including SB 53 in California, the RAISE Act in New York, and Illinois’ SB 315. In the essay, Amodei said transparency alone no longer matches the risks posed by frontier AI systems.

Amodei Flags Cybersecurity as First AI Risk to Fully Materialize

Amodei identified cybersecurity as the first risk category to have fully materialized, pointing to Claude Mythos Preview, which he said solved 73% of expert-level cyber challenges, a benchmark he said no prior AI system had cleared. 

The essay warns that frontier models could disrupt the financial sector and critical infrastructure. Anthropic shipped Claude Fable 5 on June 9 with safeguards that block high-risk cyber and biology requests, though Amodei argued that voluntary limits of that kind cannot substitute for binding rules across the industry. 

He also warned that autonomy risks may follow, writing in the essay that AI models have gone from barely being able to write a coherent line of code to writing most of the code at major AI companies. 

Amodei Proposes Wage Insurance, Weapons Ban, and Data Broker Restrictions

On economic policy, Amodei proposed wage insurance, retention tax incentives, and workforce training grants to address AI-driven job displacement. If job displacement becomes a long-term condition, Amodei said universal basic income could be financed through corporate or capital gains taxes. 

On civil liberties, Amodei called for a ban on fully autonomous weapons in domestic law enforcement and urged Congress to close the data broker loophole that enables bulk surveillance purchases.

Amodei Calls for Democratic Coalition to Control Chips and Semiconductor Equipment

Geopolitically, Amodei called for a coalition of democracies to control chips and semiconductor manufacturing equipment. He cited two pending US bills – MATCH and OVERWATCH – which he described as steps toward tighter export controls on chips and semiconductor equipment, as a foundation for broader coordinated restrictions among democratic nations. Amodei wrote that he considers public concern about AI accurate.