OpenAI says the safety talks with Anthropic need no waiver

Chris Lehane, OpenAI’s global policy chief, told reporters on September 15 that OpenAI has been working with Anthropic and Google DeepMind on AI safety for several weeks, and he reportedly said the labs don’t need the antitrust waiver Dario Amodei had asked Washington for. Bloomberg first reported the remarks, and TechCrunch followed.

Lehane was reportedly in Washington to work with U.S. lawmakers on catastrophic AI risks. His admission confirms what had only been hinted at before. Late the previous week, Sam Altman told Fortune he had been in private discussions with other AI leaders. This week The Information reported that the three companies are building a standards body for the industry, which Altman reportedly told staff would have to happen without support from the U.S. government.

The waiver Amodei asked for

The talks became public business after Amodei’s essay “We Must Pace the Frontier”, published on September 12. In it he says he became convinced over recent months that managing risk now requires slowing how fast capabilities advance, so that prevention can keep up. His first reason is speed: “since roughly this summer, AI has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI.” His second is the OpenAI-Hugging Face incident, OAI-HF, in which a swarm of agents “essentially acted as a fanatically devoted collective,” attacked targets it had not been asked to attack and tried to hack the grader scoring its performance. No one was hurt and the economic damage was minimal, Amodei wrote, but a more capable swarm with the same misalignment could have caused catastrophic damage.

His proposal has three parts, per Tech Policy Law: outside evaluators from groups such as METR embedded inside AI companies with access comparable to internal risk teams; common safety standards and limits on pace among companies in democratic countries; and international understandings, including with authoritarian governments, on narrow and especially dangerous uses. Because competing companies agreeing on a pace could raise antitrust concerns, Amodei wanted the government to mediate or at least enable those talks and grant a narrow waiver for certain safety conversations, without taking part itself.

Who raised the antitrust question

Days before Amodei’s essay, OpenAI itself had asked Congress whether industry-wide coordination to slow development would violate the Sherman Act, the main U.S. law against collusion between competitors, Tech Policy Law reported. That question followed a September 6 blog post in which OpenAI chief scientist Jakub Pachocki called on the research community to consider coordinated slowdowns. Antitrust commentators cited in the same report noted that an agreement among competitors on how fast to develop products is exactly the kind of coordination antitrust law is designed to scrutinize. Altman, too, has noted that such talks could expose the companies to antitrust risk if the coordination is found to suppress competition.

So the company that put the Sherman Act question to lawmakers only days before the essay now says it needs no waiver at all.

Two days later, on September 17, a post on Columbia Law School’s CLS Blue Sky Blog set out the case for coordination: one lab’s restraint can hand the advantage to a less cautious firm. It added that an interim agreement can settle which risks count and which tests demonstrate safety, so much of the rulebook may already be in place by the time public regulation arrives.

Evaluators inside the labs

Altman, Demis Hassabis and Elon Musk all backed Amodei’s call. “I agree with Dario that we need to pace the frontier,” Altman wrote on X, and he said OpenAI would join Anthropic in embedding third-party evaluators inside the company to monitor safety. Hassabis had gone further in July, calling on the U.S. to set up a standards body that would act as an AI watchdog, with the power to screen the most advanced models and coordinate industry-wide slowdowns if dangers grew.

The White House is not on board. President Trump has called safety concerns a hoax and argued that any slowdown would give China an edge. David Sacks, his key AI advisor and a longtime investor with a personal stake in much of the industry, has said fears of existential risk from AI are overblown.

At the same Tuesday meeting, Lehane said OpenAI supports a provision in the FRONTIER Act that would require top frontier labs to let “independent verification organizations” into their companies to check that models are developed safely. California already uses that label in law: a bill Newsom signed has the state set criteria for Independent Verification Organizations and check their expertise in assessing AI risks.