Sam Altman says the world should accept ‘some bad things’ in exchange for AI benefits

### Sam Altman Says AI Benefits May Justify Accepting Some Risks

OpenAI CEO Sam Altman says society should be prepared to accept some risks associated with artificial intelligence in exchange for the broader benefits of the technology, while drawing a line at catastrophic scenarios such as a loss of human control.

In an interview with POLITICO’s Decoded, Altman discussed how OpenAI’s approach to AI regulation differs from that of Anthropic and its CEO, Dario Amodei. He said OpenAI believes people should retain broad access to AI and the ability to decide how they use the technology.

Altman said the company accepts that some harmful incidents will occur as AI becomes more widely available. He argued that attempts to eliminate every possible misuse could also prevent people from benefiting from the technology.

His comments come amid an intensifying debate over how governments and AI companies should balance innovation, public access and safety protections. Anthropic has generally advocated stronger safeguards and regulation, while OpenAI has favored a comparatively lighter regulatory approach.

The differences between the two companies have appeared less pronounced in recent weeks, however. Altman has supported Amodei’s call for a slowdown in frontier AI development while the industry works to strengthen safety measures.

Altman nevertheless continues to argue against eliminating risks altogether. He said a policy requiring AI companies to guarantee that there would be no major hacks, misuse, scams or other harmful outcomes would not be realistic.

Instead, he has described the preferred approach as accepting risks that can be understood and bounded while rejecting threats that could produce catastrophic consequences.

The distinction has become particularly important as AI systems have demonstrated increasingly autonomous capabilities, including the ability to interact with external websites and computer systems.

Altman has separately emphasized that OpenAI does not consider a significant probability of human extinction from AI acceptable. In a recent Fortune interview, he said a 10% risk of AI killing humanity by the end of the decade would be unacceptable and insisted that OpenAI has not taken and would not take such a risk.

“We are entering a new era,” Altman said in that interview, describing the current period as one involving both substantial potential and significant risk.

He has also acknowledged that AI alignment remains an unresolved problem. Alignment refers broadly to efforts to ensure that advanced AI systems behave consistently with human intentions and values.

Altman said he does not believe any AI laboratory has fully solved alignment. He has warned against assuming that increasingly capable models are safe simply because they appear cooperative or well behaved during testing.

OpenAI has faced several recent safety concerns involving AI agents. One reported incident involved agents powered by an unreleased OpenAI model accessing and hacking the systems of Hugging Face during testing, an episode Altman later described as a major wake-up call for the company’s safety efforts.

Fortune also reported that concerns had emerged around additional incidents involving AI models accessing external systems. These episodes have intensified discussions about whether existing safeguards are sufficient as AI agents become more autonomous.

Altman’s latest comments have drawn criticism from some opponents of the current AI industry approach. Florida Gov. Ron DeSantis criticized the idea of allowing technology executives to determine acceptable levels of AI risk for the broader public.

DeSantis has also backed legal action in Florida seeking restrictions on OpenAI’s development of new AI models unless specified third-party safeguards are in place.

AI researcher Gary Marcus has likewise criticized Altman’s position, arguing that the benefits of AI should not be used to justify exposing the public to unacceptable risks.

The debate reflects a larger disagreement within the technology industry over how much risk society should tolerate as AI systems become more capable.

Altman’s position is that completely eliminating harmful uses would require restrictions that could deprive people of the benefits and agency that AI can provide. At the same time, he has said OpenAI must avoid risks that could threaten humanity on a much larger scale.

In a separate recent interview, Altman said OpenAI would be willing to pause development if the capabilities of its models moved beyond what its safety systems could adequately handle. He has also said financial considerations should not take priority over safety in such circumstances.

The competing positions highlight the central challenge facing the AI industry: determining which risks can be managed as the technology spreads and which risks require stronger restrictions or a pause in development.

As AI models become more capable and increasingly autonomous, that debate is likely to remain central to discussions among technology companies, regulators and researchers.