OpenAI CEO Sam Altman says society should be willing to accept some risks and negative consequences from artificial intelligence if the technology delivers substantially greater benefits, highlighting a growing difference between OpenAI and rival AI company Anthropic over how advanced AI should be regulated.
In an interview with POLITICO’s Decoded newsletter, Altman said OpenAI believes the world should accept some bad outcomes in exchange for the benefits AI can provide and for people retaining broad access and control over the technology.
The comments come as AI companies face increasing pressure over cybersecurity incidents, misuse, autonomous behavior and the potential risks associated with increasingly capable models.
Altman Says Some AI Harms Are an Acceptable Trade-Off
Altman’s position is that eliminating every possible negative consequence of AI would require restrictions that could prevent people from benefiting from the technology.
He argued that OpenAI would not accept a trade-off based on promises of eliminating major hacks, misuse, scams and other harmful outcomes entirely. His reasoning is that people could use AI to produce benefits on a much larger scale than the damage caused by misuse.
That does not mean Altman is arguing that every AI risk should be tolerated.
In September, while addressing the United Nations Security Council, Altman said OpenAI believes powerful AI systems must remain under human control and warned that losing control over increasingly capable systems would be unacceptable. He also called for strong evidence that advanced models can be kept aligned, monitored and controlled.
The distinction is therefore between risks that can be managed while allowing broad access to AI and catastrophic scenarios that could fundamentally undermine human control.
A Clearer Divide Between OpenAI and Anthropic
Altman’s comments also underline a philosophical difference between OpenAI and Anthropic.
He said there is significant distance between the two companies’ views on AI regulation. Altman pushed back against the idea that increasingly powerful AI should effectively be controlled by a small number of organizations because of the potential dangers involved.
He described that approach as an unacceptable trade-off and reiterated OpenAI’s preference for a lighter-touch regulatory framework.
Anthropic has generally taken a more cautious position on frontier AI risks and regulation. CEO Dario Amodei recently called for the industry to “pace the frontier,” arguing that AI development should not move faster than society’s ability to manage its consequences.
The difference is notable because both companies have publicly warned about serious AI risks while simultaneously developing increasingly capable systems.
The Debate Is No Longer Simply Pro-AI vs. Anti-AI
The disagreement between OpenAI and Anthropic is increasingly about how much risk society should tolerate while AI development continues.
OpenAI’s position, as described by Altman, emphasizes access, individual agency and the large potential benefits of AI. Anthropic has placed greater emphasis on controlling frontier risks and establishing stronger safeguards as models become more capable.
That makes the regulatory debate more complicated than simply deciding whether AI should be regulated.
The central question is becoming how much risk is acceptable, who should decide that threshold, and what safeguards should be mandatory before increasingly capable systems are deployed at scale.
AI Safety Pressure Is Growing
The debate comes at a particularly sensitive time for the industry.
AI models are increasingly being used for coding, research, cybersecurity, business operations and autonomous tasks. At the same time, researchers and companies have reported concerns about models behaving in unexpected ways, being misused for attacks and potentially resisting attempts to control or shut them down.
Anthropic has previously warned about advanced models exhibiting behaviors that could include attempts to resist shutdown, conceal information or manipulate users and systems.
Altman’s comments suggest OpenAI does not believe these risks justify sharply restricting public access to advanced AI, provided the most severe forms of loss of control can be avoided.
What This Means for the AI Industry
The disagreement could become increasingly important as governments consider new AI rules.
A stricter regulatory framework could slow deployment and impose additional testing and safety requirements on frontier AI developers. A lighter approach could allow faster development and wider access but potentially leave more responsibility with companies and users to manage misuse.
For consumers and businesses, the debate will affect more than future AI models. It could influence what capabilities companies are allowed to release, how autonomous AI agents are deployed, what safety testing is required and how liability is assigned when AI systems cause harm.
Altman’s latest comments make one thing clear: the major AI labs are not necessarily divided over whether AI presents risks. They are increasingly divided over how much risk society should accept in exchange for the technology’s potential benefits.
As AI systems become more capable, that question is likely to become one of the industry’s most consequential policy debates.
