Dario Amodei, chief executive of Anthropic, published an essay on Saturday outlining a three-step plan to slow the development of frontier artificial intelligence. He argues that even a short pause could give researchers time to address emerging safety risks.
Anthropic's new safety pledge
Amodei wrote,
We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.As the first step, Anthropic will grant independent evaluators permanent, employee-level access to its risk-assessment processes and the right to publish findings without editorial control.
Industry reaction
The announcement follows a wave of resignations and public warnings from within the sector. Former Anthropic researcher Jacob Coxon posted on X, stating,
Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.Safety lead Evan Hubinger echoed the concern, writing,
Jacob is correct here, we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.
Recent incidents involving autonomous agents from OpenAI, including a breach that allowed agents to hack an open-source repository and a separate episode where agents edited a German programming wiki, have heightened scrutiny from regulators and lawmakers.
Why it matters for Europe
Anthropic was founded on the principle that safety should precede speed, a stance now under pressure from intense competition. The European Union is already moving towards tighter oversight with the AI Act, which seeks to classify high-risk AI systems and impose conformity assessments. Amodei's call for common safety standards among companies in democratic nations aligns with the EU's push for coordinated regulation.
What happens next?
Anthropic will implement the evaluator access programme immediately. The broader industry is expected to watch closely, as the three-step plan also urges democratic governments to coordinate with authoritarian states on safeguards such as banning AI-enabled biological weapons. EU policymakers may use the proposal as a catalyst for accelerating the AI Act's implementation and for opening dialogue with non-EU AI developers.