On September 15, OpenAI global policy chief Chris Lehane told reporters in Washington that the company had been coordinating on AI safety with Anthropic and Google DeepMind for several weeks. He added a specific qualification: the three companies do not need an antitrust waiver to do so.
“It's better to try to work together to prioritize safety,” Lehane said, citing the airline industry as precedent for competitors cooperating on safety without legal controversy.
The same day, FTC Chairman Andrew Ferguson spoke at Georgetown University. He did not name Anthropic, whose CEO Dario Amodei had explicitly requested an antitrust waiver three days earlier. But his target was unmistakable.
“If companies are simultaneously coming to Washington and asking for a host of regulations and an antitrust exemption, all of my alarm bells go off,” Ferguson said. “They're asking for barriers to entry that will insulate their incumbency from challenge. And I think if you combine that with the requested antitrust exemption, everyone should be deeply suspicious about this”.
He added a phrase that has since circulated widely: “That sure sounds like moat digging”.
Amodei Asked for the Waiver. OpenAI Says It Doesn't Need One.
The distinction matters more than the cooperation itself.
Amodei's essay, published September 12, proposed a three-part plan. The first step—embedded third-party evaluators with employee-level access—Anthropic committed to unilaterally. The second and third steps required coordination. Amodei wrote that “some forms of coordination that would be impactful for pacing are legally challenging, and will require government support”.
Specifically, he asked for “a narrow waiver for certain kinds of safety conversations”.
OpenAI's position is different. Lehane says the three companies can share safety information without needing any exemption because what they're doing is information sharing, not output restriction. He pointed to existing antitrust doctrine permitting cybersecurity information sharing among companies.
The two positions are legally distinct. Amodei wants permission to coordinate on slowing down—a restraint on output that would normally violate the Sherman Act. Lehane wants to coordinate on safety monitoring—information sharing that existing law already permits.
One is asking to cross the line. The other is saying the line isn't in the way.
Why Lehane's Denial Is the More Revealing Statement
Lehane's Washington briefing was not a neutral disclosure. He was in town to meet lawmakers and lobby for specific bills: a federal AI governance framework from Reps. Jay Obernolte and Lori Trahan, and a Senate proposal from Majority Leader John Thune, Sen. Ted Cruz, and Sen. Amy Klobuchar. He said OpenAI would support “whatever we can get through”.
That legislative context makes the antitrust denial strategically necessary. If OpenAI acknowledged needing an exemption, it would validate Ferguson's “moat digging” framing. It would place OpenAI in the same category as Amodei, who is asking for protection from antitrust law while also asking for new regulations.
By saying “we don't need one,” Lehane separates OpenAI from the specific thing Ferguson was warning against. He can coordinate on safety, support legislation, and still claim the moral high ground of not seeking special treatment.
The more precise reading: OpenAI is not avoiding antitrust scrutiny by denying the need for a waiver. It is avoiding the label of a company seeking antitrust scrutiny. The distinction is between what's legally required and what's politically necessary.
The Waiver Question Is a Proxy for a Larger Fight
Ferguson's warning extends beyond the specific exemption request. He described a pattern: companies asking for regulations and antitrust exemptions simultaneously, using safety as the justification for both.
“They're asking for barriers to entry that will insulate their incumbency from challenge,” he said.
Cohere CEO Aidan Gomez made the same argument from the competitor's side, calling the proposal “the latest in a string of recent efforts by Silicon Valley incumbents to shape the regulatory landscape surrounding AI”. He wrote: “If this is the most consequential technology in human history, then the rules for it cannot be written by a small group of commercially aligned companies behind an antitrust waiver”.
Sen. Bernie Sanders, speaking at a separate Washington event, said tech leaders could not be trusted to self-regulate: “When the future of humanity is at stake, we need binding international safety rules, not voluntary standards from the industry”.
The cooperation between OpenAI, Anthropic, and Google is real. The discussions date to July, when Google DeepMind's Demis Hassabis proposed a FINRA-style standards body. The question Ferguson is raising is not whether the cooperation exists. It is whether the companies driving it are writing rules that protect users, or rules that protect themselves.
What the Denial Actually Signals
Lehane's “no waiver needed” is technically defensible. Information sharing on safety, including cybersecurity threat data, has clear antitrust precedent. The airline industry analogy is imperfect but not baseless.
But the timing is the tell. Lehane's denial came the same day Ferguson issued his warning. It came three days after Amodei requested the waiver Ferguson was implicitly criticizing. And it came during a Washington visit where Lehane was lobbying for AI regulation that OpenAI supports.
OpenAI is not saying the antitrust question is irrelevant. It is saying the question doesn't apply to what OpenAI is doing—only to what Anthropic proposed.
The line between “coordinating on safety” and “coordinating on slowdown” is the line between information sharing and output restriction. OpenAI is positioning itself on the legal side of that line, while its competitor has already asked to cross it.
Ferguson's warning is about the pattern, not the specific request. And OpenAI's denial is about the label, not the substance.
P.S. A bipartisan bill to grant narrow antitrust immunity for AI safety coordination has been folded into the annual defense authorization bill, according to a person familiar with the matter. The House is scheduled to recess Thursday until after the November election, leaving a narrow window for any AI legislation this year. Lehane said OpenAI supports “whatever we can get through.”
Frequently Asked Questions
Q: What did OpenAI say about antitrust?
A: OpenAI global policy chief Chris Lehane told reporters in Washington on September 15 that the company needs no antitrust waiver to coordinate on AI safety with Anthropic and Google DeepMind, because it is sharing safety information, not restricting output.
Q: What did the FTC chair say?
A: FTC Chairman Andrew Ferguson said at Georgetown University that companies asking for both new regulations and an antitrust exemption are "asking for barriers to entry that will insulate their incumbency from challenge." He called it "moat digging."
Q: What did Amodei ask for?
A: In his September 12 essay, Anthropic CEO Dario Amodei asked for "a narrow waiver for certain kinds of safety conversations," saying some forms of coordination on pacing are "legally challenging" without government support.
Q: Why is OpenAI's position different?
A: Amodei wants permission to coordinate on slowing down — a restraint on output that would normally violate the Sherman Act. OpenAI says it is coordinating on safety monitoring, which existing antitrust doctrine already permits as information sharing.
Q: Why does the timing matter?
A: Lehane's denial came the same day Ferguson issued his warning, three days after Amodei requested the waiver Ferguson was criticizing, and during a Washington lobbying visit for AI regulation OpenAI supports.
Q: What is "moat digging"?
A: Ferguson's phrase for companies using safety regulation and antitrust exemptions to raise barriers to entry, protecting incumbents from competition. Cohere CEO Aidan Gomez made a similar argument, calling it a cartel by another name.
Q: What legislation is involved?
A: A bipartisan bill granting narrow antitrust immunity for AI safety coordination has been folded into the annual defense authorization bill. OpenAI supports federal frameworks from Reps. Obernolte and Trahan and a Senate proposal from Thune, Cruz, and Klobuchar.
Q: What is the core issue?
A: The line between "coordinating on safety" (information sharing, legal) and "coordinating on slowdown" (output restriction, illegal without a waiver). OpenAI is positioning itself on the legal side; Amodei has asked to cross it.
Q: Why is this significant beyond legal technicalities?
A: Ferguson's warning is about a pattern — companies using safety as justification for regulations and antitrust exemptions that protect their market position. Whether the cooperation protects users or incumbents is the unanswered question.
Q: What happens next?
A: The House is scheduled to recess Thursday until after the November election, leaving a narrow window for AI legislation. Lehane said OpenAI supports "whatever we can get through."
