Does AI Need an Antitrust Exemption to Avoid Killing Everyone?
Former DOJ antitrust chief Jonathan Kanter on AI slowdowns, competition, and cartels.
By Nilay Patel
As artificial intelligence systems grow more capable—and more deeply embedded in critical infrastructure, healthcare, finance, and national security—a provocative question has moved from science fiction into serious policy debate: should AI developers be shielded from antitrust enforcement in the name of safety?
The argument, advanced by some AI labs and safety advocates, goes roughly like this: if frontier AI poses existential risks, then competing developers should be allowed to coordinate on safety standards, share model internals, and slow down releases without running afoul of competition law. In other words, an antitrust exemption could function as a safety valve.
But critics—including former Department of Justice antitrust chief Jonathan Kanter—warn that this framing inverts the real problem. Competition, not consolidation, is what keeps powerful technologies accountable.
The Case Against an Antitrust Exemption
Kanter, who led the DOJ's Antitrust Division from 2021 to 2024, has argued that exempting AI companies from antitrust scrutiny would concentrate power in the hands of a few dominant firms precisely when oversight is most needed. A handful of companies already control the compute, data, and talent pipelines that define frontier AI.
An exemption, he cautions, could:
- Enable cartel-like coordination among incumbent labs, freezing out smaller competitors and open-source efforts.
- Entrench market concentration in AI chips, cloud infrastructure, and foundation models.
- Weaken the feedback loop between competition and safety, where rival firms challenge each other's claims and practices.
Kanter's position reflects a broader shift in antitrust thinking: that competition policy is itself a safety mechanism. When no single actor can dictate terms, the risks of misuse, negligence, or unchecked capability growth are distributed and contested rather than centralized.
AI Slowdowns and the Coordination Problem
The debate over "AI slowdowns" cuts to the heart of the tension. Safety researchers sometimes advocate for coordinated pauses or compute caps. But such measures, if implemented by private firms without regulatory oversight, could resemble market allocation—the very behavior antitrust law was designed to prevent.
A more defensible path, Kanter and others suggest, is for governments to set safety rules directly, rather than delegating that authority to a self-selected club of AI developers. Public regulation, after all, is subject to public accountability. A private cartel is not.
The 2026 Policy Landscape
As of 2026, the debate has sharpened. The EU's AI Act is in phased implementation, the U.S. has pursued a sectoral approach combining antitrust enforcement with executive guidance on frontier model reporting, and several states have introduced their own AI competition bills. Meanwhile, major AI labs continue to lobby for safe-harbor provisions that would let them collaborate on safety without triggering antitrust liability.
Regulators, however, remain skeptical. The Federal Trade Commission and DOJ have signaled that they view AI markets as a priority enforcement area, and the concept of a blanket antitrust exemption remains politically unlikely—especially given bipartisan concern about tech concentration.
The Bottom Line
The question in the headline—does AI need an antitrust exemption so it doesn't kill everyone?—may be deliberately provocative, but the underlying issue is real. The answer from competition enforcers like Kanter is clear: no. If AI is dangerous enough to warrant extraordinary coordination, that coordination should come through transparent, democratically accountable regulation—not through private arrangements that shield dominant firms from competition.
In this view, antitrust law is not an obstacle to AI safety. It is a precondition for it.
This article is based on commentary and interviews featuring Jonathan Kanter and reflects ongoing policy debates as of 2026.
via The Verge AI