Anthropic and OpenAI Differ Over Responsibility for Halting Risky AI
Anthropic and OpenAI have diverged over who should be responsible for pulling the plug on AI systems judged to pose unacceptable risks. The disagreement centers on whether responsibility for halting advanced models should rest with the companies that develop them, public authorities, or another form of oversight.
The contrasting positions frame a broader governance challenge: the developers of increasingly capable AI systems are also best placed to recognize certain dangers, but independent intervention may be needed when commercial interests conflict with safety concerns. The available report does not detail the organizations’ specific proposals, leaving open how they differ in practice or what mechanisms either supports for managing extreme risks.
As AI development advances, the issue remains a test of how responsibility, technical control, and external oversight should be divided. The divergence between Anthropic and OpenAI suggests that major AI labs do not yet share a common answer.