OpenAI CEO Sam Altman said developers should defer training AI models unless they can make an “extremely strong case” that people can keep them under control, warning that advanced systems could strain institutions and concentrate power.
At a Sept. 23 U.N. Security Council briefing on AI and international security, Altman warned that systems could advance faster than institutions can adapt. He said they might shift power toward a smaller group and make decisions people no longer understand or control.
Altman said commercial rivalry was no excuse for accepting excessive technological risk. He also argued that corporate choices should remain subject to democratic decision-making.
ANI reported Nov. 4 that Altman wrote on X, “People should always be in control of the future.” ANI did not provide a link to the post.
In a June essay co-written with OpenAI Chief Scientist Jakub Pachocki, Altman said people should retain a role in steering AI and resolving competing priorities, with human judgment still needed as the technology develops. The essay also recognized that commercial and national rivalries continue to exert pressure on AI development.
Anthropic CEO Dario Amodei, writing in September, advocated a slower pace for frontier AI and ongoing safety assessments by independent evaluators.
The National reported a similar message from Altman at an ADIPEC event in Abu Dhabi on Nov. 2. EconoTimes previously covered his remarks at ADIPEC.
OpenAI said Nov. 2 that Cathedral Systems researchers would join its next internal security sprint. The researchers are to work with OpenAI teams and use OpenAI models to test defenses and identify weaknesses. The announcement described security testing but did not say the sprint would assess whether a model could remain under human control.
