A widening argument among AI executives has turned a shared acknowledgement of risk into a sharper question: should companies pause to build safeguards, or press ahead while managing problems as they arise?

Anthropic chief executive Dario Amodei has urged the industry to slow down, comparing the moment to rival carmakers reviewing their own safety records after a serious brake failure. Rather than treating an incident as a chance to attack a competitor, he said, companies should ask whether their own systems are sufficiently secure: “We may not have had this big, high-profile incident, but I’m sure we’re not perfect.”¹ He has proposed independent evaluators inside AI firms, safety coordination among democratic countries and, ultimately, wider international cooperation.¹

Nvidia chief executive Jensen Huang takes the opposite view on pace. He argues that new regulation is unnecessary and that companies should decide not to release systems when they cannot establish their safety. His formulation — “Run as fast as you can” — couples acceleration with a duty to pause when an institution loses control of its product.¹ Huang also rejects a trade-off between innovation and safety, describing safety as an engineering problem rather than a reason for a blanket slowdown.²

OpenAI’s Sam Altman occupies a more qualified position. He has said the public is “right to be afraid” as AI capabilities advance and warned that failures are easier to imagine than before.² At the same time, he has argued that companies can keep safety ahead of capability growth — and should “slow down or stop” if they cannot.²

The three leaders therefore converge on the need for safeguards, but differ over their enforcement. Amodei stresses independent scrutiny and coordinated standards; Huang places responsibility chiefly with companies; Altman calls for tougher security and voluntary collaboration. Critics of self-regulation, including Anthropic’s Jack Clark, have warned that leaving the sector wholly unregulated would mean “rolling dice with immense risks.”²