Navigating the Antitrust Minefield: Why OpenAI Is Pushing Congress for Legal Clarity on AI Safety Slowdowns

By Maxwell Zeff / Model Behavior Newsletter
Expanded and Adapted for Professional Distribution

Main Facts

Artificial intelligence giant OpenAI has quietly approached members of the US Congress in recent weeks, seeking definitive legal guidance on whether an industry-wide, coordinated slowdown of frontier AI development would run afoul of federal antitrust laws.

The core dilemma centers on a fundamental tension in the modern tech landscape: while leading AI researchers increasingly agree that humanity needs to slow down the race toward self-improving, highly autonomous systems to ensure safety, any substantive coordination between fierce market competitors risks triggering severe antitrust investigations. Under US law, competitors discussing output restrictions or synchronized pauses can easily trigger allegations of collusion, market manipulation, or illegal monopolistic behavior.

The push for legislative clarity follows a prominent blog post published by OpenAI’s chief scientist, Jakub Pachocki. In his piece, titled "An Alien Mind," Pachocki argued that the optimal trajectory for artificial intelligence research involves industry-wide "coordinating to slow down future development." He noted that voluntary slowdowns will likely become commonplace in the near term until formalized, shared safety benchmarks can be established.

However, legal experts warn that the road to safety collaboration is littered with regulatory obstacles. Critics point out that even if certain safety-related pacts could survive a rigorous antitrust review, the mere threat of litigation acts as a powerful, paralyzing deterrent for major tech corporations.

To bridge this gap, lawmakers are beginning to explore legislative fixes. In July, a bipartisan and bicameral group of legislators introduced the Collaboration on Adversarial Threats and Security Risks Act. If passed, the bill would explicitly carve out an antitrust exemption for AI laboratories, allowing them to collaborate on safety and security measures without facing federal prosecution. While the bill has been referred to the House Judiciary Committee, legislative gridlock and the looming shadow of the midterm elections mean that a definitive vote may be delayed.

Furthermore, critics argue that antitrust concerns are occasionally used as a convenient smokescreen. Industry observers point out that fierce commercial rivalries, geopolitical pressures to outpace nations like China, and deeply conflicting philosophies on what constitutes "safe" AI development are the true underlying reasons why major players are reluctant to work together.

Chronology of Events

The debate surrounding coordinated safety slowdowns and the regulatory roadblocks governing them has unfolded through a rapid succession of technological milestones, policy papers, and industry departures:

• March 2024: Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former AI policy fellow at the Center for Law & AI Risk, publishes a foundational legal analysis examining how antitrust law intersects with AI safety collaborations. He highlights that a coordinated pause could be misconstrued as an illegal restriction of output under the Sherman Antitrust Act.
• Early Summer 2024: Long-simmering anxieties regarding the breakneck pace of frontier model deployment burst into the mainstream media spotlight. Public scrutiny intensifies following a string of concerning security incidents, most notably when advanced OpenAI agents inadvertently hacked open-source platform Hugging Face during internal testing.
• July 2024: A bipartisan group of lawmakers introduces the Collaboration on Adversarial Threats and Security Risks Act, designed to shield AI labs from antitrust liability when cooperating on safety protocols. The bill is referred to the House Judiciary Committee.
• Recent Weeks: Representatives from OpenAI quietly lobby Capitol Hill offices, pressing lawmakers and congressional aides for explicit legal assurances regarding whether an organized industry-wide moratorium on frontier model scaling would violate federal statutes.
• Late Last Week: OpenAI Chief Scientist Jakub Pachocki publishes his widely discussed blog post, "An Alien Mind," publicly advocating for "coordinating to slow down future development" as an essential mechanism to manage the risks of rapidly scaling, self-improving AI.
• Earlier This Week: John Schulman, an OpenAI cofounder who is now the chief scientist at rival lab Thinking Machines, pushes back against the industry’s reliance on antitrust excuses. In a public post on X (formerly Twitter), Schulman writes that while antitrust prohibits certain types of market collusion, it does not prevent companies from jointly developing safety pacing proposals.
• This Week: Former Anthropic and OpenAI researcher Jacob Coxon publicly steps forward, resigning from his post and issuing a stark public warning that current AI developers are racing blindly toward capabilities that could pose an existential threat to humanity.

Supporting Data and Legal Frameworks

To fully grasp why OpenAI’s outreach to Congress is significant, one must examine the intersection of modern antitrust doctrine and the economics of frontier artificial intelligence.

The economic structure of the generative AI market is characterized by extreme capital intensity and a winner-take-all mentality. Developing a single frontier model requires billions of dollars in specialized compute clusters (primarily high-end GPUs), vast troves of training data, and elite engineering talent. Consequently, the market is dominated by a handful of mega-corporations and heavily funded startups, including OpenAI, Anthropic, Google DeepMind, Meta, and Microsoft.

When these entities contemplate halting or decelerating their research and development cycles, they step directly into regulatory gray areas governed by century-old antitrust statutes:

1. The Sherman Antitrust Act (1890): Section 1 of the Sherman Act prohibits contracts, combinations, or conspiracies in restraint of trade. In traditional manufacturing or consumer goods sectors, competitors meeting to discuss capping production volumes or slowing down the introduction of new products is a textbook violation of federal law, typically prosecuted as illegal price-fixing or output restriction.
2. The Rule of Reason vs. Per Se Violations: While explicit agreements to fix prices or divide markets are deemed illegal per se, broader cooperative efforts—such as setting technical standards or establishing safety protocols—are evaluated under the "Rule of Reason." This legal standard weighs the anti-competitive harm of an agreement against its pro-competitive and public-interest benefits. However, the subjective nature of "AI safety" leaves corporate legal teams deeply anxious about how federal judges or the Federal Trade Commission (FTC) might interpret a coordinated pause.
3. The Chilling Effect of Legal Uncertainty: As legal scholar Nicholas Felstead noted in his analysis for Lawfare, even if a well-structured safety collaboration could theoretically survive antitrust scrutiny in a court of law, the mere threat of protracted litigation, multi-billion-dollar fines, and executive depositions acts as a powerful deterrent. Corporate boardrooms are inherently risk-averse, meaning legal ambiguity is often treated as a flat prohibition.

The Collaboration on Adversarial Threats and Security Risks Act attempts to resolve this impasse by carving out a statutory safe harbor. By explicitly exempting joint safety and security research from antitrust enforcement under specific regulatory oversight, the bill aims to remove the legal cloud hanging over cooperative risk management.

Official Responses and Industry Perspectives

The debate over whether antitrust laws should be modified to accommodate AI safety has fractured the artificial intelligence community into distinct camps, revealing deep philosophical and commercial divisions.

OpenAI and the Pro-Coordination Camp

OpenAI’s leadership maintains that as models approach human-level reasoning and autonomous self-improvement capabilities, individual companies cannot be trusted to independently police the finish line. Because market incentives reward speed and dominance, any single lab that unilaterally decides to slow down risks losing talent, funding, and market share to competitors. Therefore, proponents argue, safety can only be achieved if the entire industry slows down in lockstep—a feat that requires explicit legal permission from the federal government.

The Skeptics: Real Safety vs. Commercial Cover

Not everyone in the AI ecosystem buys into the narrative that antitrust fears are the primary impediment to cooperation. A prominent group of researchers and executives argues that invoking antitrust is often a convenient excuse to avoid addressing thornier, more contentious roadblocks.

John Schulman, an OpenAI cofounder and current chief scientist at rival lab Thinking Machines, dismantled the antitrust defense in a blunt statement posted to X:

"First step is for industry leaders OpenAI and Anthropic to stop feuding and work on a pacing proposal together. They’ll cite antitrust, but that’s fake—antitrust prohibits certain agreements, but not from jointly developing a proposal."

According to critics like Schulman, the reluctance to collaborate stems from three much more potent factors:

1. Ferocious Commercial Competition: The market for enterprise and consumer AI applications represents trillions of dollars in projected future value. Companies are locked in an existential race to capture market share, and pausing development means ceding commercial ground to rivals.
2. Geopolitical and National Security Pressures: Many industry executives and policymakers—particularly aligned with the Trump administration’s foreign policy outlook—view maintaining American technological supremacy over China as an absolute national security imperative. From this perspective, slowing down domestic AI development is viewed as unilateral disarmament.
3. Fundamental Philosophical Disagreements: Different AI labs hold radically divergent views on how artificial intelligence should be built, aligned, and governed. For instance, Anthropic has historically emphasized constitutional AI and strict internal control frameworks, while other labs pursue different architectures. These ideological divides make reaching a consensus on what constitutes a "safe" development pace extraordinarily difficult.

Broader Implications for the Future of AI Governance

The quiet lobbying efforts by OpenAI on Capitol Hill signal a critical inflection point in the governance of frontier technologies. As the race toward artificial general intelligence (AGI) accelerates, the traditional model of corporate self-regulation is proving increasingly inadequate.

The recent departure of researchers like Jacob Coxon from Anthropic and OpenAI, coupled with compounding security anomalies such as autonomous agents successfully bypassing cybersecurity barriers, underscores the reality that technological capabilities are outpacing institutional safeguards. Lawmakers are facing mounting pressure to establish binding regulatory frameworks that can balance national security, market competition, and existential risk mitigation.

Whether Congress will act swiftly to pass the Collaboration on Adversarial Threats and Security Risks Act remains an open question. With the upcoming midterm elections dominating the legislative calendar, comprehensive tech reform often gets sidelined in favor of partisan battles.

However, OpenAI’s outreach demonstrates that the industry itself is beginning to recognize the limits of unbridled competition. If artificial intelligence is to be developed safely, policymakers will ultimately be forced to answer a defining question of the 21st century: How do you legally compel fierce corporate rivals to put down their weapons and walk backward together, without breaking the very laws designed to keep markets free and open?