Former DOJ Antitrust Chief Says AI Firms Need Rules, Not a Cartel Deal

Former Justice Department antitrust chief Jonathan Kanter says the biggest AI companies do not need an antitrust exemption to build safer products, and he argues Congress should instead create clear rules for liability, security and competition. His comments, made in a wide-ranging interview as pressure grows around AI safety and regulation, land at a moment when tech giants are asking for more coordination while critics warn that such requests could shield incumbents and weaken competition.

Kanter, who led major antitrust cases against Google, Apple and Ticketmaster during the Biden administration and now teaches at Washington University and Carnegie Mellon, described the AI sector as a powerful new technology running without basic traffic rules. In his view, companies should be responsible for what their systems do, and the government should focus on setting boundaries that reduce harm without letting industry coordination become a disguised cartel.

Why this AI regulation fight matters now

The debate over AI safety has become a debate over market power, liability and industrial policy all at once. Leaders at major model labs have warned publicly that frontier systems may pose serious risks, while also pressing for policy solutions that would make it easier for firms to coordinate on safety, testing and development pace.

That combination has set off alarm bells among antitrust watchers. If competitors are allowed to align on how fast they move, what safeguards they adopt or how much they spend, those agreements could reduce rivalry and freeze out smaller players, including open-weight model developers and foreign competitors.

Kanter’s central argument is that the law should distinguish between legitimate safety collaboration and behavior that softens competition. He said firms can share threat information or build common repositories for malicious activity without needing special permission to do so. But he warned that an agreement among rivals to “slow down” could easily cross into unlawful coordination.

What did Jonathan Kanter say about AI companies and cartels?

He said the strongest case for an AI industry-wide slowdown is that companies feel overwhelmed by the pace of innovation and want help drawing “lines on the road.” But he also laid out a more cynical interpretation: that money-losing labs may be looking for a way to reduce costly race conditions before they go public or face harsher investor scrutiny.

Kanter did not accept the idea that a safety push automatically equals cartel behavior. At the same time, he said the antitrust risks are real if firms use safety as a pretext to dampen competition or lock in the position of the largest players.

Kanter argued that companies do not need to coordinate with rivals to build products that are safe and secure, and said the better analogy is ordinary product responsibility: if a company’s technology causes harm, the company should answer for it.

How he compares AI to cars, planes and other industries

To explain his position, Kanter used analogies from transportation and manufacturing. If an aircraft has a safety failure, he said, the answer is not for rival airlines or plane makers to pause innovation together; the answer is to fix the underlying design and production problem. The same logic, he argued, should apply if an AI system behaves dangerously.

He said the goal should not be to create a permission structure for competitors to move in lockstep. Instead, he wants a framework that forces companies to build systems that are secure by design and imposes consequences when those systems fail.

• Companies should be liable for harmful outputs or actions by their systems.
• Safety collaboration should be allowed when it involves threat-sharing or security coordination.
• Competitors should not be allowed to use safety concerns to justify blanket restraint of trade.
• Regulation should set baseline duties without turning the market into a managed cartel.

How would liability work if an AI agent causes harm?

In Kanter’s view, the answer is straightforward in principle: firms should be accountable when their tools or agents act on their behalf and cause damage. He argued that it should not matter whether the harmful actor is a human employee, a software tool or an AI agent.

That position pushes the debate toward product liability, tort law and clearer federal standards. If a consumer deploys a company’s agent and the agent hacks another system, steals data or carries out an unauthorized attack, Kanter believes the company should not be able to wash its hands of responsibility simply because the misbehavior was automated.

He acknowledged that courts and lawmakers may need to clarify exactly how such liability works in practice. But he said the basic principle already exists in other areas of law: companies are responsible for the tools they deploy and the harm those tools cause.

What “robot jail” means in this debate

Kanter used the phrase “robot jail” to describe a possible remedy for dangerous AI systems: the idea that a harmful agent could be taken offline or removed from the market. The term is colorful, but the legal concept behind it is familiar — regulators and courts can force products to be modified, restricted or withdrawn when they are unsafe.

That approach, he suggested, would be more effective than waiting years for a case to crawl through the courts. It would also keep the focus on the product and the company, rather than on abstract debates about whether AI is too novel to fit existing law.

Why does Kanter think Congress has fallen behind?

Because, he said, lawmakers have not kept up with the speed or scale of the technology. Kanter argued that Washington has become too slow to set baseline safety rules, even as tech platforms have accumulated enormous influence over daily life, commerce and information.

He placed some of the blame on the broader political system, pointing to the effects of gerrymandering and campaign-finance rules as structural obstacles to reform. In his telling, Congress has often been unable to act decisively while major companies have lobbied against rules they publicly say they want.

He also suggested that the absence of clear federal rules has shifted more of the burden onto courts, states and existing liability doctrines. That, he said, is a workaround — not a substitute for legislation.

How AI regulation compares with the social media era

One of Kanter’s core warnings is that the public has already seen a version of this movie before. He pointed to the long delay before regulators and courts forced major social media companies to confront harms tied to design choices, youth engagement and mental health.

According to his argument, the AI debate is happening earlier because trust in Big Tech has already eroded. That matters: lawmakers and the public are now more suspicious when companies ask for flexibility, especially if they simultaneously resist limits in practice.

Kanter also contrasted the legal environment around today’s AI companies with the immunity platforms enjoyed under Section 230 and the more permissive attitudes of the earlier platform era. He said those conditions made it easier for companies to scale quickly without full accountability for downstream harms.

Kanter said the lesson of the social media era is simple: if a technology is likely to cause serious harm and the company sees that coming, it should not wait for the law to catch up before changing course.

What role do the states play?

The states have become a key pressure point because they have moved faster than Washington. Kanter said state governments have often been more willing to test rules for AI, online safety and platform accountability than Congress has been to adopt national standards.

That has created another fight: some lawmakers and industry groups have pushed for a federal moratorium that would block states from setting their own rules for years, even in the absence of a strong federal framework. Kanter viewed that as backwards — especially if Congress is unwilling to offer a meaningful replacement.

In practice, that means states may remain the main source of near-term AI constraints, whether through consumer-protection laws, privacy rules, product-liability theories or targeted safety legislation.

What this means for OpenAI, Anthropic, Google DeepMind and others

The companies at the center of the frontier AI race are now operating under growing pressure from multiple directions. They want to avoid an unmanaged race to the bottom on safety, but they also want room to compete, spend aggressively and move fast enough to keep up with rivals.

Kanter’s comments underline a difficult reality for them: asking for policy help may be understandable, but the solution is unlikely to be a blanket immunity deal. If lawmakers or regulators decide these companies need guardrails, those guardrails may come with stricter liability rather than a cooperative carve-out.

That would make the market harder, not easier, for incumbents. It could also favor firms that can prove their systems are safer, not simply larger or better funded.

Timeline of the policy battle

The current clash over AI does not exist in a vacuum. It sits on top of years of antitrust enforcement, platform litigation and a broader political backlash against Big Tech.

Who is Jonathan Kanter?

Kanter is one of the most prominent antitrust lawyers in Washington. As assistant attorney general for the DOJ’s Antitrust Division under President Joe Biden, he helped spearhead a more aggressive stance toward market concentration, especially in digital platforms and other highly consolidated sectors.

He now teaches law at Washington University and technology policy at Carnegie Mellon, and he continues to shape the public conversation on how competition law should respond to emerging technologies. In the AI debate, that makes him a particularly influential voice: he combines a regulator’s instincts with a lawyer’s skepticism of broad corporate claims.

His remarks suggest that the coming fight will not simply be about whether AI is dangerous. It will also be about who gets to decide how the industry is organized, how much coordination is allowed and whether safety becomes a genuine public good or a convenient excuse for incumbent power.

What comes next?

The next phase of the debate is likely to be more concrete. Lawmakers will face pressure to decide whether existing law is enough, whether new legislation is needed and whether companies should bear explicit liability when their models or agents cause harm.

At the same time, the antitrust question will remain central. If the biggest AI firms continue asking for exceptions, safe harbors or coordinated action, critics will keep asking whether those requests protect the public — or protect the market positions of the companies making them.

Kanter’s answer is that the distinction matters enormously. Safety rules are not the same thing as a permission slip for competitors to act together. In the AI era, he argues, the government should draw the lines on the road, but it should not hand the keys of regulation to the industry that wants the fewest limits.

That tension — between real safety needs and the risk of cartel-like coordination — is now one of the defining policy fights in artificial intelligence. And it is unlikely to go away soon.