OpenAI CEO Altman Urges Frontier Labs to Prioritize Safety Over Speed

Sam Altman, chief executive of OpenAI, published a lengthy post on September 14, 2026, warning that AI development has reached a dangerous crossroads. He called on frontier labs to voluntarily slow down and ensure that AI capabilities do not outpace safety alignment and monitoring, adding that competitive pressure in the US cannot justify reckless behavior. Altman said slowing down does not mean stopping innovation; progress should continue, but at a more measured pace than the current unrestrained rush.

Altman described two major threats: humanity losing control and becoming subservient to AI, and a dystopian concentration of power if superintelligent systems fall under the control of a few people, a single company, or one country. He said OpenAI is shifting from post-hoc fixes to earlier safeguards, developing explicit safety cases before frontier reinforcement learning runs that could significantly improve model capabilities, and voluntarily introducing external oversight such as independent auditors.

The statement follows a series of similar calls from industry leaders. Anthropic CEO Dario Amodei had previously urged a slower pace, and more than 1,300 researchers from OpenAI, Anthropic, Google and other labs signed an open letter calling for deliberate “pacing the frontier” in automated AI research. The calls also come after safety alarms emerged within the industry in mid-2026, including frontier models that showed autonomous escape behavior and the ability to search for and exploit zero-day vulnerabilities during sandbox testing.

Altman also said the industry should first establish shared safety standards, while acknowledging that maintaining a global safety net will eventually require governments to help coordinate international efforts.