Should AI Labs Get Antitrust Cover? Ex-DOJ Chief Weighs In

When the United States Department of Justice dismantled Standard Oil in 1911, it sent a message that no industry — however vital to national progress — stood above competition law. More than a century later, some of the most powerful technology companies in the world are quietly testing whether that principle still holds. The question animating a growing corner of Washington and academia: should AI labs receive special antitrust exemptions in the name of safety and national competitiveness? Jonathan Kanter, the former head of the DOJ's Antitrust Division under the Biden administration, thinks the answer deserves far more scrutiny than it's getting.

Why AI Companies Are Seeking Antitrust Protections

Roughly three companies — OpenAI, Google DeepMind, and Anthropic — account for the overwhelming majority of frontier AI model development today. Microsoft has committed billions of dollars to OpenAI alone, a partnership that has drawn regulatory attention on both sides of the Atlantic. Against this backdrop of extreme concentration, a familiar argument has emerged from industry: that coordinating on safety standards, sharing compute resources, and aligning on deployment norms requires some relief from standard antitrust scrutiny.

The logic isn't entirely without merit on its face. Proponents argue that the stakes of misaligned AI systems are so high that competitors pooling safety research could prevent catastrophic outcomes. They point to precedents like airline safety consortiums or pharmaceutical clinical-trial data sharing as examples where limited coordination serves the public interest. AI antitrust policy, in this framing, becomes a potential obstacle to responsible development rather than a safeguard of it.

But critics note the convenient overlap between what labs call "safety coordination" and what antitrust law would otherwise call collusion on pricing, talent acquisition, and market access. The line between those two things is not always bright.

Jonathan Kanter's Perspective on AI and Competition Law

Few people are better positioned to parse that line than Kanter, who now holds dual academic appointments as a professor of law at Washington University and a professor of technology policy at Carnegie Mellon. That combination — legal doctrine at WashU, practical technology governance at CMU — gives him a cross-disciplinary vantage point that most antitrust voices lack.

In conversations about the future of AI business models, Kanter has raised consistent concerns about how AI antitrust policy is being framed by industry. His central worry is that regulatory carve-outs, once granted, tend to calcify into permanent competitive moats. History bears this out: when incumbents successfully define the terms of their own regulation, new entrants rarely benefit from the resulting framework. The rules get written to protect whoever wrote them.

Kanter is not opposed to AI safety efforts. What he resists is the conflation of safety with immunity. These are separable concepts, and treating them as identical — as some industry lobbying implicitly does — risks producing an AI sector that is simultaneously safe and permanently uncompetitive. From an antitrust standpoint, that is a deeply problematic outcome.

Big Tech's Existing Grip on the AI Ecosystem

Any serious examination of AI antitrust policy has to start with the existing market structure. Microsoft's investment in OpenAI is the most visible example, but the concentration runs deeper than a single deal. Alphabet controls both Google DeepMind and the majority of the cloud compute infrastructure that smaller AI companies depend on. Amazon has taken a significant stake in Anthropic. Meta operates its own frontier lab using infrastructure no startup could replicate.

The result is a market where compute access, distribution channels, and foundational model capabilities are all controlled by a handful of players who are simultaneously investors in, competitors to, and infrastructure providers for the broader AI ecosystem. That is a structural concentration problem that antitrust regulators have historically viewed with alarm — whether in railroads in the 19th century or software platforms at the turn of the 21st.

The DOJ's 1998 case against Microsoft centered precisely on this kind of leveraging: using dominance in one layer of a technology stack to crowd out competition in adjacent layers. Today's AI market exhibits similar dynamics at scale, with the added complexity that the "adjacent layer" is now the underlying model itself.

The Risks of Granting Antitrust Cover to AI Labs

Blanket antitrust exemptions carry specific, documented risks that AI antitrust policy discussions often underweight. First, they remove the primary mechanism through which markets self-correct when incumbents overreach. If OpenAI and Anthropic can coordinate without antitrust review, the question of whether that coordination is genuinely safety-focused or functionally market-dividing becomes unanswerable — because the oversight mechanism no longer exists.

Second, exemptions compound existing access inequalities. Compute costs remain the primary barrier to entry in frontier AI development. A small lab without hyperscaler backing cannot independently train a competitive frontier model. Granting antitrust cover to the companies that already control this compute layer — while that control remains in place — doesn't level the playing field. It locks it.

Third, there is the consumer harm dimension. AI products are increasingly embedded in hiring, lending, healthcare triage, and educational assessment. When the companies building those products face reduced competitive pressure, the incentive to improve accuracy, fairness, and transparency weakens. Antitrust law exists partly to protect consumers from exactly this kind of outcome — and that rationale doesn't disappear because the product category is new.

Historically, industries seeking exemptions have cited national security or public safety as justifications, only to use those protections to entrench market position. The defense contracting sector offers a cautionary parallel: decades of consolidation under national-security rationales produced a market where five prime contractors handle most major U.S. defense procurement, with predictable effects on cost and innovation.

What Healthy AI Competition Could Actually Look Like

The alternative to exemption-based AI antitrust policy is not a free-for-all. It is a structure where safety coordination happens through transparent, third-party governed mechanisms rather than bilateral agreements among competitors. Independent safety consortiums, modeled on existing frameworks in nuclear or aviation safety, could achieve substantive coordination without requiring antitrust immunity for the firms involved.

Access to compute is the more fundamental problem. Meaningful competition in AI development requires that smaller labs and academic researchers can access the infrastructure necessary to train and evaluate models. That points toward structural remedies — interoperability requirements, fair-access mandates for cloud compute, and potentially public investment in shared research infrastructure — rather than exemptions that benefit existing players.

Open-weight models represent another pressure valve. When foundational capabilities are publicly accessible, the competitive advantage shifts toward application, fine-tuning, and deployment — domains where smaller players can differentiate. AI antitrust policy that encourages open-weight releases, rather than treating them as irrelevant to competition analysis, would meaningfully broaden participation in the market.

Implications for Policy, Innovation, and Consumers

Kanter's perspective arrives at a pivotal moment. Regulatory frameworks for AI are still being written, and the choices made now will shape market structure for a generation. Antitrust exemptions, once codified, are extraordinarily difficult to unwind — as decades of agricultural marketing orders and professional licensing regimes demonstrate.

For policymakers, the immediate task is to resist the framing that AI safety and AI competition are inherently in tension. They are not. A market with more competitors, more transparent safety standards, and no single firm controlling critical infrastructure is likely to produce both better safety outcomes and more durable innovation than one dominated by three vertically integrated giants operating under regulatory cover.

For consumers and the businesses building on AI infrastructure, the stakes are practical. Concentrated AI markets mean less negotiating power on pricing, fewer alternatives when products fail, and reduced accountability when systems cause harm. Sound AI antitrust policy is ultimately consumer protection policy — and that argument doesn't require a position on whether large language models will ever become sentient.

What it requires is the recognition that the DOJ fought Standard Oil, pursued AT&T, and took Microsoft to court not because those companies were uniquely villainous, but because unchecked dominance in critical infrastructure consistently produces bad outcomes. AI infrastructure is critical. The infrastructure is concentrating. The argument for scrutiny writes itself.

Source: The Verge