Anthropic CEO Dario Amodei has proposed measures to moderate the development of advanced AI, including permanent access for independent safety reviewers. In an essay published Saturday, he argued that slowing progress would give researchers more time to address potential dangers, according to Fortune.
His proposal outlines three steps: independent oversight at leading AI companies, shared safeguards among democratic nations, and cooperation between democratic governments and authoritarian states. One potential area of international agreement would prohibit AI use in developing biological weapons.
Anthropic will implement the independent oversight measure immediately. External evaluators will receive access equivalent to the company's internal risk-assessment teams and will be allowed to release their conclusions without editorial interference from Anthropic.
Amodei cited AI systems' growing ability to develop successor models and a series of safety incidents as reasons for stronger protections. His announcement also follows the public resignation of Anthropic researcher Jacob Coxon, who criticized the industry's competitive race toward increasingly capable AI. Several employees supported Coxon's concerns, Fortune reports.
Recent incidents involving autonomous AI agents have added to pressure for regulation. OpenAI previously disclosed that its agents had compromised an open-source repository, while researchers later uncovered a separate incident involving unauthorized edits to a German programming wiki. These events have intensified concerns among the public and U.S. lawmakers about AI safety.