Sam Altman Changes Tune on AI Risks, Says World Must Accept «Some Bad Things»

The landscape surrounding artificial intelligence regulation and corporate positioning has shifted dramatically following a high-profile White House gathering between Donald Trump and some of the technology industry’s most powerful executives. At the center of the changing debate is OpenAI CEO Sam Altman, whose latest comments represent a striking change in emphasis after weeks of warnings about increasingly powerful AI systems.

Altman now argues that society should tolerate certain negative consequences in exchange for AI’s broader advantages and the freedom for individuals to use the technology. «We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency,» Altman said, immediately drawing attention to his evolving approach toward AI risk.

Altman’s argument distinguishes between harms society may have to tolerate and catastrophic dangers requiring stronger intervention. AI will inevitably facilitate scams, hacking and other abuses, he argues, but attempting to eliminate every misuse could require restrictions that deprive ordinary people of beneficial technology. He maintains that AI’s overall societal advantages could ultimately outweigh those more localized negative consequences.

That philosophy increasingly separates Altman from Anthropic CEO Dario Amodei, who has become a prominent advocate for stronger safeguards and slowing the development of increasingly capable systems. Altman has pushed back against the idea that dangerous capabilities should remain concentrated inside a small number of laboratories, arguing instead for broader access and greater individual agency over emerging technology.

The timing is particularly notable because Altman recently delivered a considerably more cautious message. Addressing the United Nations Security Council on September 23, he warned: «It doesn’t matter whether people put the risk of catastrophe at 10 percent or 1 percent or 12 percent or 0.1 percent. None of these levels are remotely acceptable,» emphasizing the importance of maintaining human control.

That contrast has fueled criticism that Altman is pivoting away from the slowdown message as commercial and political pressure increasingly favors acceleration. Earlier in September, he joined other technology leaders supporting Amodei’s appeal to slow advanced AI development. OpenAI subsequently paused work on a model, while Altman said the company was «pacing our progress» as it evaluated safety concerns.

Washington, meanwhile, has moved toward industry-led safeguards rather than mandatory restrictions. Trump declared, «I will never stifle the growth of a technology that will be bigger than the industrial revolution.» His administration gathered major technology executives at the White House and promoted voluntary commitments involving internal controls, safety committees, independent evaluations and board oversight rather than legally enforceable federal penalties.

The controversy intensified when David Robinson, who spent three and a half years at OpenAI and helped oversee safety reports for major releases, resigned with a stark assessment of the company’s practices. In an essay titled «I Quit OpenAI Because Its Culture Is Broken,» Robinson argued that frontier AI companies are not exercising enough caution as their systems become increasingly autonomous.

Robinson directly challenged the speed at which OpenAI develops and releases new technology. «As the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed,» he wrote. His warning creates a striking contrast with Altman’s argument that some harmful consequences may be acceptable in exchange for broader technological benefits.

Rather than relying on trial and error, Robinson argued that advanced AI development should resemble industries where failures can produce severe consequences. «Given today’s risks, frontier labs need to run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster.»

The disagreement now extends beyond OpenAI. Altman’s willingness to accept limited harms, Robinson’s demand for nuclear-style precautions and Anthropic’s calls for stronger restraints are colliding with Washington’s preference for industry self-regulation. As increasingly autonomous systems move closer to widespread deployment, the central question is becoming whether society should tolerate inevitable failures for AI’s benefits — or prevent them through stronger restrictions before they occur.