OpenAI chief executive Sam Altman has argued that society should tolerate certain harms from artificial intelligence in exchange for much larger gains. In a recent interview with Politico’s Decoded, he said the world ought to accept some adverse outcomes because people will generate orders of magnitude more beneficial uses than harmful ones.
He listed hacks, scams, and other forms of misuse as costs that should not be eliminated at the price of stalling the technology or concentrating control in a handful of labs.
Altman framed the stance as a deliberate contrast with Anthropic. He rejected the idea that a single San Francisco laboratory should hold advanced systems and ration their benefits solely to prevent every failure.
In his view, heavy restrictions aimed at perfect safety could themselves concentrate power in a few companies, individuals, or countries—an outcome he described as another path to serious harm.
He has also said he supports tougher rules in specific high-risk domains and has previously signaled agreement with calls for more careful pacing of frontier systems, while still favoring a lighter overall regulatory approach.
OpenAI, he indicated, does not see a serious loss of human control over AI as a plausible near-term prospect, though he has urged caution before rushing into that future.
The remarks landed days after an OpenAI safety leader stepped down and described the company’s current trajectory as unacceptable.
Critics on professional networks pushed back sharply. Some argued that the upside tends to concentrate among a small set of firms and investors while the downside falls on the public, small businesses, and institutions that lack comparable resources.
Others noted that the scale of failure changes when systems move into medicine, industrial controls, or critical infrastructure, where an error can affect physical safety rather than only data or money.
Several commentators stressed that neither the pace of deployment nor the distribution of risk is fixed; democratic rules, liability standards, and design choices remain available levers.
Those comments sit inside a wider debate over whether development should slow.
In September, Anthropic chief executive Dario Amodei urged the industry to ease the pace of capability gains so safety work could catch up, warning that an unchecked race could produce catastrophic results.
Altman and Elon Musk publicly aligned with the broad concern at the time, while stressing that any pause would need to account for competitors.
President Donald Trump rejected the slowdown.
He said the main safeguard required is capable national leadership, argued that existing criminal and regulatory authority already covers the companies, and repeated that whoever leads in AI will hold a decisive advantage.
He has treated warnings of existential risk as overstated and has framed efforts to restrain data centers or model progress as beneficial mainly to rivals.
Beijing has likewise dismissed Western slowdown appeals as fearmongering intended to hobble a competitor.
Chinese officials have continued to back domestic model builders and reduce reliance on foreign technology, even as President Xi Jinping has spoken of keeping development under human control and preventing loss of control.
Other major economies remain split between industrial policy aimed at catching or matching US and Chinese capacity and more restrictive rules focused on high-risk uses. The result is a competitive environment in which calls for coordinated restraint have so far found little traction among the governments most able to enforce them.