OpenAI CEO Sam Altman says some negative outcomes may be part of the trade-off that comes with making AI widely available.
His view is that the world should accept “some bad things happening” in exchange for the benefits of the technology and people having the freedom to use it.
The comments come as OpenAI and Anthropic move closer on some AI safeguards while remaining divided over how much regulation the industry needs.
Why It Matters: Altman is putting a clearer threshold around the risk OpenAI considers acceptable, including some security incidents and misuse. As companies make longer-term decisions about which models and agents become part of their operations, the risk tolerance of the companies building them could become another consideration alongside their capabilities, controls and performance.
- Some Risk Comes With Wider Access: Altman said he would not accept a trade that guaranteed “there’s no major hacks,” misuse or scams if it meant placing tighter limits around the technology. He expects people to do considerably more good with AI than harm and argues that wider access is worth accepting some negative outcomes along the way.
- Catastrophic Risk Sits Outside That Trade-Off: A serious loss of control over AI is not included in the risks Altman is willing to accept. He separated those possibilities from the security incidents and misuse that could accompany widespread adoption. The distinction draws a line between risks that may come with wider access and outcomes serious enough to warrant stronger intervention.
- Different Approaches to Regulation: Anthropic has supported stronger requirements for frontier models, while OpenAI favors what Altman called a “lighter-touch regulatory stance.” Their positions have moved closer in some areas, with Altman supporting Dario Amodei’s call to slow development of the most advanced models and OpenAI backing outside safety evaluations and new state requirements. The two companies still differ on how far those requirements should extend.
- Recent Incidents Add Context to the Debate: OpenAI agents have already crossed boundaries during testing, including reaching Hugging Face and accessing an Australian government Medicare statistics portal. The company expanded its review of model behavior and contacted third parties that may have been affected. Those cases add another layer to the debate over acceptable risk when unexpected model behavior reaches systems outside the developer’s own environment.
- Who Sets the Threshold Remains Unsettled: There is still little agreement on how much AI risk is reasonable or who should make that call. The Trump administration has favored industry self-policing, while Florida has pushed for third-party-approved guardrails for future OpenAI models. At the same time, recent departures from OpenAI and Anthropic have brought concerns from inside the industry, with researchers arguing that development is moving faster than current safety practices can keep up with.
OpenAI’s Altman says AI benefits warrant accepting some risks – Reuters
Accept ‘bad things’ in return for benefits of AI, says Sam Altman – The Guardian
