OpenAI CEO Sam Altman issued a stark warning on September 14 that advanced artificial intelligence could produce two severely damaging outcomes: humanity losing control over increasingly capable systems, or excessive power becoming concentrated in a single company, person, country, or research laboratory. His statement shifts the AI debate beyond computing performance toward the question of who will control frontier technology as its influence grows.
Altman endorsed Anthropic CEO Dario Amodei’s proposal to pace frontier AI development while researchers strengthen safety protections. He clarified that pacing does not mean stopping technological progress. Instead, laboratories should accept slower development when safety reviews require more time, even when those interventions carry high costs.
OpenAI is now introducing safety cases before selected reinforcement learning training runs, particularly when researchers expect a training run to produce significant capability gains. The company is also maintaining safety work before releasing completed models, placing evaluations at both the development and deployment stages.
Altman called for industry-wide action without waiting for legislation. He said competitive pressure within the United States should never justify reckless development, and that model capabilities must not advance beyond alignment and monitoring systems. He supported common standards covering model misalignment, monitoring, and safety, welcomed independent auditors, and backed a federal framework establishing consistent requirements for frontier developers.
Governments, he said, should play a larger role in coordinating international safety rules. Altman also stressed that OpenAI remains committed to ensuring AI serves humanity, while warning that without adequate supervision and shared standards, more capable systems could progress dangerously.
