AI Safety Deals Face Antitrust Threat as U.S.-China Race Accelerates

WASHINGTON — OpenAI has asked members of Congress whether an agreement among competing artificial intelligence developers to slow technological rollouts could violate U.S. antitrust statutes, according to a WIRED report citing sources familiar with the discussions. The inquiry has made federal antitrust laws a central legal hurdle for companies seeking to coordinate safety standards and has prompted bipartisan action on Capitol Hill.

Section 1 of the Sherman Antitrust Act of 1890 subjects horizontal agreements between direct competitors that restrict output, delay product rollouts, or limit commercial activity to strict scrutiny by the U.S. Department of Justice (DOJ) Antitrust Division and the Federal Trade Commission (FTC). Without explicit legal protections, joint efforts by competing technology firms to establish binding development pauses could be construed as unlawful restraints on trade.

Sens. Adam Schiff (D-Calif.) and Jim Banks (R-Ind.) have introduced legislation intended to create a regulatory safe harbor. The proposed bill would allow AI developers to participate in certain security and safety collaborations without facing federal antitrust liability if they provide advance notice of their agreements to the Department of Justice.

The legislative effort comes as OpenAI Chief Scientist Jakub Pachocki urges AI developers to adopt voluntary slowdowns until companies can verifiably demonstrate that advanced systems operate safely. Market dynamics, however, create strong disincentives against voluntary, unilateral pauses, according to tech policy experts and industry observers.

“Commercial and geopolitical competition in the AI space is incredibly intense, leading to a concerning dynamic where companies are incentivized to release products before their risks are fully understood,” said Miranda Bogen, chief technologist at the Center for Democracy and Technology. “Even when internal staff knows more research and testing is needed, their companies are facing immense pressure to cut corners and skip critical safety tests, despite evidence piling up about the consequences of moving too fast.”

The leading laboratories in the race include OpenAI, Anthropic, Google DeepMind, and Meta. Capital investment requirements and compounding technological advances add to the pressure. “Every advance under current conditions yields many millions or billions more in funding and puts the creators of that advance in a greater position of power and influence,” said Duncan Sabien, head of communications at the Machine Intelligence Research Institute (MIRI).

Sabien said companies such as OpenAI and Anthropic face persistent pressure to maintain development velocity because technological progress feeds directly into subsequent iterations. “This is especially true since gains in intelligence are compounding; each new system makes it easier to train and deploy the next, more-advanced system,” he said. “Sans some sort of coordination mechanism, stepping back just means the other guy gets a lead.”

Those pressures have affected internal corporate policies. In February, OpenAI and Anthropic both walked back certain previously stated safety commitments. Jared Kaplan, chief science officer at Anthropic, said that slowing development unilaterally offered little practical value while rival firms continued advancing their frontier models.

Internal security pauses have nonetheless occurred. In August, OpenAI halted internal work on its Project Astra capabilities after red-teaming and risk evaluations revealed cybersecurity vulnerabilities requiring stronger technical safeguards before further testing could proceed.

The conflict between rapid commercial deployment and existential risk mitigation has also led to departures from major AI laboratories. Jacob Coxon, a former engineer at Anthropic, publicly announced his resignation, citing concerns about the pace of AI development and its potential catastrophic risks to humanity.

“The people building AI earnestly believe that it could kill us all by the end of the decade,” Coxon wrote in a public statement on X, emphasizing that his concerns were not a marketing tactic. “If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible—but I hear the same people express fear privately. No other human activity poses this level of danger.”

Sabien said individual departures rarely change corporate trajectories unless broad coordination occurs. “If enough of them achieve common knowledge that they should all stop, then they can break through the coordination barrier and stop together rather than just being replaced by the next slightly-less-cautious person,” he said.

Industry-wide slowdowns are also complicated by macroeconomic and geopolitical rivalries, especially the competition between the United States and China for dominance in artificial intelligence infrastructure and capability. National security considerations have repeatedly influenced regulatory policy within the executive branch.

In May, President Donald Trump delayed an executive order on artificial intelligence because of concerns that strict oversight frameworks could impede American technological momentum and cede ground to Chinese developers. Trump later signed a revised executive order in June, creating a voluntary review process through which federal agencies can evaluate advanced frontier models before commercial release.