Anthropic CEO Warns AI Is Getting Too Powerful—While Releasing Powerful AI
Anthropic CEO Dario Amodei called for binding U.S. safety rules for frontier AI models, arguing that transparency alone is no longer enough as AI advances too quickly for existing policy. He compared regulation to the FAA model: advanced systems should face mandatory third-party testing, auditing, secure handling of model weights, incident reporting, and government power to block unsafe deployments. He also urged governments to prepare for AI-driven job displacement, limit surveillance and autonomous weapons, and deepen coordination among democratic countries on critical AI technology. The essay comes as Anthropic expands its restricted cybersecurity/government model and public-facing safeguards, both of which have drawn scrutiny over misuse risks and user restrictions. Amodei framed his stance as a public-safety issue, while critics, including OpenAI’s Sam Altman, have argued that AI safety rhetoric can be used to concentrate control.
