Jonathan Kanter Says AI Safety Work Does Not Need an Antitrust Exemption

The debate over whether rival AI labs need legal cover to coordinate on safety has a new rebuttal: Jonathan Kanter says they do not. In a newly published interview, the former head of the Justice Department’s Antitrust Division argued that companies can build safer systems and share focused security information without an antitrust exemption—and should be responsible when the AI agents they deploy cause harm.

Kanter’s comments arrive as frontier AI companies including OpenAI, Anthropic, Google DeepMind and Elon Musk have been described as calling for a slower development pace or government intervention. The Verge’s Nilay Patel asked whether those appeals reflect a need for outside rules to resolve a race between competitors, or something closer to industry coordination that could raise antitrust concerns.

A safety argument is not a free pass

Kanter offered both a charitable and a skeptical reading of the slowdown push. Companies may genuinely fear the consequences of powerful systems developing without clear rules, he said. But he also raised the possibility that firms under heavy spending and competitive pressure could prefer an industry-wide slowdown that gives them room to improve their economics before potential public offerings.

These companies do not need to coordinate in order to deliver safe and secure products to society.

Jonathan Kanter, former head of the DOJ Antitrust Division, speaking to The Verge

A narrow lane for cooperation

His distinction is between sharing concrete safety and security information and agreeing to compete less aggressively. Kanter said a clearinghouse or repository for information about threats, malicious bots or similar risks is the kind of limited collaboration that can help companies make safer products. In his view, antitrust law generally does not prohibit that work, so a special exemption is unnecessary.

The line Kanter draws

• Share specific information about threats and malicious bots through a clearinghouse or repository.
• Do not treat a shared desire to slow competition as necessary for building safe products.

His proposed consequence: liability

Kanter’s answer puts the immediate burden on the companies making and deploying the systems. He argued that a company could face liability when its AI agent hacks or damages someone else’s technology, comparing such agents with human employees acting on a company’s behalf. The question is not only whether an AI product malfunctioned, but whether a company should answer for what its deployed agent does.

He said existing product-liability frameworks may sometimes apply when technology causes harm, but also acknowledged the need for government action to clarify safety standards and consequences. That leaves a divided job: firms must build safe and secure products now, while government sets clearer rules for failures that reach the public.

The unresolved move belongs to policymakers

Kanter rejected the idea that the absence of new legislation excuses companies from releasing products they believe could cause severe harm or break into other systems. His position does not settle how liability would work in a particular AI case. It does sharpen the policy choice: lawmakers can define the rules of the road, but companies cannot outsource their own safety obligations to a future exemption or statute.