Lawmakers Push Back on Amodei’s Call for an AI-Safety Antitrust Waiver

AI companies seeking permission to cooperate on safety face a pointed objection: some lawmakers and a former top antitrust enforcer say they already can. In September 30 reporting by CNBC, Sen. Elizabeth Warren and former Justice Department antitrust chief Jonathan Kanter rejected the need for the waiver proposed by Anthropic CEO Dario Amodei.

Amodei’s essay, “We Must Pace the Frontier,” urged AI companies to voluntarily establish shared standards. He argued that the U.S. government should mediate or enable those discussions through a narrow waiver for certain safety conversations. The proposal also contemplated collective limits on unchecked AI progress, which Amodei acknowledged could pose legal challenges.

Sharing threats is not the same as sharing a slowdown

Kanter drew the legal boundary around what competitors agree to do. Companies can exchange information about malicious threats and develop standards for safer products without an exemption, he told CNBC. His objection concerned agreements that would stop companies from innovating or competing, rather than discussions about making their systems safer.

He said companies do not need to make slowdown decisions jointly. He also warned that safety could become a pretext for other anticompetitive coordination. That warning concerns what an exemption might permit; it is not a finding that the companies have already used safety discussions to suppress competition.

Warren likewise said existing antitrust law already contains an exception for companies working together on safety. Sen. Richard Blumenthal said companies do not need a waiver to prevent catastrophic consequences when cooperating under government oversight. Republican Sen. John Hoeven rejected a blanket exemption, citing lawmakers’ dissatisfaction with earlier technology-industry carveouts.

A bill would protect more than conversations

The waiver has congressional supporters, too. Reps. George Whitesides and Bob Latta, along with Sens. Adam Schiff and Jim Banks, carry a bipartisan bill that would create a carveout. CNBC’s description of the bill identifies two forms of protected cooperation:

• Sharing information about AI security threats to prevent, investigate or mitigate risks.
• Coordinating to reduce AI security risks by delaying or limiting deployment of high-risk models.

Rep. Sam Liccardo, who pushed for Whitesides’ bill, said any protection must be narrowly focused on testing, evaluation and training that make models safer. He argued that intense competition gives companies little incentive to collaborate across many other activities. His defense puts the scope of permitted cooperation at the center of the proposal.

Kanter’s opposition to a waiver does not amount to opposition to AI regulation. He separately urged Congress to consider basic rules protecting the public from harmful products and services. His proposed path is government-set safeguards without giving competitors additional permission to make joint business decisions.