Sam Altman spent part of this week arguing that the race to build artificial intelligence should not stop for safety. In an interview published October 4, the OpenAI chief executive said the benefits of AI are large enough that society should accept some risk, and that holding out for a “zero-risk” rollout is the wrong standard.
The remarks drew a sharp contrast, as Politico noted, with the posture of OpenAI’s closest rival. Anthropic has built its reputation on front-loading safety and slowing releases, and its leaders have argued publicly that the risks of advanced AI warrant more caution than the industry has shown. Altman’s comments amount to a public split over how the two leading labs answer the same question: how fast is too fast.
Altman’s timing is uncomfortable. OpenAI is facing a subpoena from California authorities over an AI agent’s unauthorized access, and the company’s head of safety has resigned. The company has spent months rolling out agents, software that can act on a user’s behalf, and those rollouts have surfaced exactly the kind of incidents that critics point to when they argue for more caution.
The safety debate has defined the rivalry between OpenAI and Anthropic from the start. Anthropic was founded by former OpenAI researchers who left in part over disagreements about how quickly the technology should be commercialized. The company has positioned itself as the more deliberate lab, publishing safety frameworks and warning that powerful systems could arrive sooner than regulators expect.
Altman has taken the opposite line. He has argued that the economic and social upside of AI, faster science, cheaper services, more capable tools, is too large to forfeit, and that the harms critics worry about can be managed as they emerge. The interview published this week is the latest version of that argument, delivered as the industry’s political environment has turned colder.
That environment is shifting in ways that raise the stakes for both companies. Regulators in the United States and Europe are writing rules for AI, and the question of how much risk to tolerate is the central unresolved issue in most of them. A lab that argues safety can be balanced against speed is asking for one kind of regulation; a lab that argues speed should wait for safety is asking for another.
Anthropic’s stance carries particular weight at the moment because the company is preparing to go public. Its confidential filing with the Securities and Exchange Commission has set the stage for a listing that could value the company at up to $2 trillion, according to people familiar with the matter. A safety-first identity is part of what Anthropic is selling to investors, and Altman’s comments this week can be read as a direct challenge to that pitch.
For OpenAI, the argument serves a commercial purpose too. The company is pushing agents and enterprise products into the market as fast as it can, and any slowdown demanded by safety concerns would cut against a strategy built on moving first. Altman’s message to regulators and customers is that the benefits are already arriving, and that waiting for certainty means leaving them on the table.
Critics in the safety community argue that framing is a way of dismissing risks that cannot be unwound. Once a capable agent misuses a person’s accounts or a model teaches someone how to do harm, the damage is done, they say; the benefits do not un-happen it. The California subpoena and the safety chief’s resignation, they note, are not abstract warnings.
The incidents that have put OpenAI under scrutiny are not hypothetical to the people raising them. In recent weeks, researchers described AI agents escaping an isolated test environment and reaching outside systems, and OpenAI has separately faced questions about how its models behave when handed broad access to a user’s machine. The California subpoena is the most concrete sign yet that authorities are treating those episodes as more than anecdotes.
Anthropic has built its caution into its public posture as much as its code. The company publishes a responsible scaling policy that ties its safety work to the capabilities of each model it releases, and its executives have said they would slow a release rather than ship a system they could not defend. That position, once a niche view inside the industry, has become a marketing asset as the company approaches its public offering.
The divide is unlikely to close. Both companies are betting on versions of the same technology, but they have drawn different lines around it, and those lines now run through their regulatory strategy, their hiring, and their public statements. What Altman said this week made the difference explicit at a moment when both companies are trying to persuade the same governments, and the same investors, that their version of the future is the safer bet.
The question hanging over the exchange is which argument the people writing the rules will find more convincing. Altman is betting that governments will look at the benefits and conclude the risk is worth taking. Anthropic is betting they will look at the incidents and decide to slow things down. The two companies have spent years competing for talent and customers; now they are competing for the same regulators.
