In a Politico interview published on 4 October 2026, OpenAI chief executive Sam Altman said his company and Anthropic remain divided over AI regulation. He argued that the benefits of the technology and people’s freedom to use it justify accepting some harms. The distinction comes despite OpenAI’s support for safety measures that Anthropic has also endorsed.
Key points
- Altman says OpenAI and Anthropic differ over how much AI-related harm society should accept in return for benefits and access.
- OpenAI has backed tougher state safety laws and its lobbyists have endorsed a bipartisan House proposal for outside evaluators.
- Altman also agreed with Anthropic chief executive Dario Amodei’s call to slow development of the most advanced models.
- Altman draws a boundary at risks involving what he calls “a serious loss of control to AI”.
Altman distinguishes harm from loss of control
Asked where his position differed from Anthropic’s and its chief executive, Dario Amodei’s, Altman said there was “a lot of daylight” between them. “We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency,” he told Politico’s Decoded. His answer put public access alongside the benefits he expects from AI, rather than treating every possible harmful use as a reason to restrict it.
Altman argued against concentrating access to powerful AI among a small number of developers on safety grounds, Seeking Alpha reported. He rejected the idea that preventing misuse would justify tightly limiting access and argued that beneficial uses would vastly outweigh harmful ones. Those are Altman’s expectations about how the technology will be used, rather than results of the safety measures now under discussion.
His tolerance for risk has a limit. Altman said he does not accept risks involving “a serious loss of control to AI”, Seeking Alpha reported. That distinction leaves room for restrictions aimed at the behaviour of the most advanced systems while preserving his argument for broad access to AI. It also makes the kind of harm a safeguard is intended to prevent central to the disagreement he describes with Anthropic.
Anthropic has been regarded within the industry as a particularly strong advocate of AI regulation. Its position has generally favoured stronger safeguards for the most advanced models, while OpenAI had favoured a lighter approach. Altman’s account of a continuing divide therefore concerns more than whether either company will support a particular bill: it concerns how each approaches the costs of restricting access and the risks of leaving it open.
OpenAI backs state laws and outside evaluators
On several measures, the companies have moved closer. Altman agreed with Amodei’s September call to slow development of the most advanced AI models. OpenAI has also followed Anthropic in endorsing new state laws with stricter safety requirements than it had previously supported, Politico reported. An agreement on those measures sits beside Altman’s argument that society should accept some harms for the sake of access and benefits.
OpenAI lobbyists have endorsed a bipartisan House proposal that would require leading AI companies to bring in outside safety evaluators. It remains a proposal, rather than an obligation already imposed on developers. The requirement, if adopted as described, would place an assessment of their systems with evaluators outside the companies developing them. Neither Altman’s stated preference for broad access nor Anthropic’s preference for safeguards settles how such an assessment would affect a model’s release.
OpenAI’s shift towards tougher rules has occurred while it has dealt with prominent hacking incidents and unexpected behaviour by its models, Politico reported. Altman nevertheless describes a philosophical difference with Anthropic after the companies have come to support some of the same measures. His agreement with slowing advanced development is particularly consequential to that description: it concerns the pace of work on the systems at the centre of their safety debate, not merely how an existing product is tested.
The proposed evaluator requirement and the state laws would work differently. One would bring outside assessors into leading companies’ safety work if the House proposal became law. The other consists of state requirements OpenAI and Anthropic have endorsed. Support for both allows a developer to favour tighter scrutiny of advanced systems while continuing to argue over how widely people should be able to use AI and which risks warrant a restriction.
Anthropic contests the cost of its safeguards
Cost is part of the dispute over who gets to set those limits. Safety rules and independent testing could increase development costs and, if expensive enough to comply with, favour well-funded incumbents over smaller companies, Seeking Alpha reported. That possibility matters to Altman’s objection to access being concentrated among a few developers. It also raises a separate question for rules supported by both companies: which developers would have to comply?
An Anthropic spokesperson disputed the suggestion that its preferred regulations would entrench the largest developers and said its proposals apply only to frontier models. Amodei has argued that Anthropic favours policies that put greater burdens on leading AI companies while allowing smaller competitors more room to develop. Anthropic’s answer therefore rests on where a rule draws its boundary, rather than on an objection to imposing costs on the companies developing the most capable systems.
Altman’s warning about loss of control and Amodei’s call to slow advanced development place both executives’ attention on the most capable AI systems. Their dispute over access persists alongside that overlap. For lawmakers considering the House proposal, the stated plan is to require top AI companies to embed outside safety evaluators; Anthropic’s spokesperson says its own regulatory proposals apply only to frontier models.
