Former US antitrust chief Jonathan Kanter says frontier AI labs do not need an antitrust exemption to cooperate on safety. In a Decoder interview, he argues that companies can share threat intelligence lawfully, while any pact to slow competition could trigger the very rules they want waived.
Watch Desk analysis
What happened
OpenAI, Anthropic, Google DeepMind and other leading AI companies have been discussing ways to coordinate as concern grows about increasingly capable systems. Kanter’s answer is not to forbid cooperation, but to distinguish practical safety work from an agreement that softens competition.
In the Decoder interview, Kanter says shared repositories for threats, malicious bots and similar security information can resemble lawful collaboration in other industries. By contrast, companies agreeing that they are competing too hard and should collectively slow down could raise antitrust problems.
What changes
- Safety coordination is not automatically a cartel
Labs can share concrete threat and security information without receiving a broad exemption from competition law. - Collective slowdown is a different proposition
An agreement among dominant companies to reduce competitive pressure could itself attract antitrust scrutiny. - Responsibility stays with the developer
Kanter argues that companies cannot blame fast-moving technology when their products or agents cause preventable harm. - Government still has a job
He wants enforceable safety standards and consequences, not an industry honour system decorated with reassuring nouns.
Why it matters
The distinction cuts through a muddled debate. AI labs do not need permission to improve their own safeguards or exchange narrowly defined threat information. What they cannot assume is that invoking existential risk makes every form of coordination benign.
That matters for smaller competitors and open-weight developers too. A sweeping exemption designed by the largest labs could shape who gets to compete, which systems reach the market and whose definition of safety becomes law.
Our read
Kanter offers a useful rule of thumb: collaborate on identifiable hazards, compete on products, and make companies answer for the systems they release. “Safety” should not become a velvet rope around the frontier club, but antitrust anxiety is no excuse for labs to ignore vulnerabilities they can address today.
Policymakers should demand specificity. If companies want legal protection for coordination, they should identify the exact information, tests or response mechanisms they need to share and explain why existing law blocks them.
What to watch
- Whether frontier labs publish the scope and governance of their current safety talks.
- Whether any proposal seeks a narrow safe harbour or a broad antitrust exemption.
- How liability is assigned when an autonomous agent causes damage or commits an offence.
- Whether smaller and open-weight developers receive a meaningful role in setting safety rules.
Discussion spark: Where should the legal line sit between useful AI safety coordination and an agreement that protects dominant labs from competition?
Sources and evidence
- Does AI need an antitrust exemption so it doesn&8217;t kill everyone???? (19 September 2026, 14:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.