OpenAI has asked members of Congress whether rival AI companies could legally agree to slow development, a question that cuts straight into the tension between safety coordination and antitrust law.
According to WIRED, citing people close to the company, OpenAI in recent weeks asked lawmakers whether an agreement among AI companies to slow development could violate antitrust rules. The outreach came as OpenAI Chief Scientist Jakub Pachocki pushed for voluntary slowdowns until developers can do a better job of showing their systems are safe.
The backdrop is a sector where companies are still expected to move quickly even when their own researchers have serious concerns.
Safety arguments collide with market pressure
Miranda Bogen, chief technologist at the Center for Democracy and Technology, told Decrypt that commercial and geopolitical competition in AI is so intense that companies are rewarded for releasing products before their risks are fully understood.
She said internal staff may know more testing and research are needed, yet firms still face heavy pressure to cut corners and skip critical safety checks, even as evidence keeps building about the consequences of moving too fast.
A pact to slow AI development could itself trigger antitrust concerns because it may be seen as a restriction on competition. Sens. Adam Schiff and Jim Banks have introduced legislation that would protect certain security collaborations, provided the Justice Department receives advance notice.
U.S.-China rivalry adds another layer
The fight over whether AI development should slow down is unfolding while the United States and China compete for leadership in the technology. U.S. President Donald Trump delayed an AI executive order in May because of concerns that it could weaken America’s lead over China. He later signed the order in June, setting up a voluntary review process for advanced models before release.
National security is only part of the picture. Financial incentives may be an even more direct obstacle to any agreement.
Duncan Sabien, head of communications at the Machine Intelligence Research Institute, told Decrypt that every major advance under current conditions can bring many millions or even billions of dollars in added funding, while also giving the creators of that advance more power and influence.
Sabien described the arms-race dynamic as real. In his view, OpenAI and Anthropic are both pushed to continue advancing despite uncertain risks because each new system helps build a stronger successor. If either company slows down on its own, without a shared arrangement, the other could pull ahead.
As he put it, gains in intelligence are compounding: each new system makes it easier to train and deploy the next, more advanced one. Without some kind of coordination mechanism, stepping back means giving someone else the lead.
Recent moves by OpenAI and Anthropic
In August, OpenAI paused internal Astra work that lacked stronger safeguards, citing cybersecurity concerns. Even so, both OpenAI and Anthropic relaxed their safety commitments in February. Anthropic Chief Science Officer Jared Kaplan argued at the time that slowing development independently made less sense if competitors were still moving ahead.
Resignation puts the debate back in focus
The argument over slowing AI development picked up again this week after former Anthropic engineer Jacob Coxon publicly announced his resignation, saying continued AI development poses risks to humanity.
On X, Coxon wrote that the people building AI earnestly believe it could kill everyone by the end of the decade, adding that this was not a "marketing stunt." He also said many executives and senior researchers may use more measured language in public, while expressing the same fear in private, and argued that no other human activity carries that level of danger.
Sabien said researchers may not accomplish much by leaving if somebody else simply replaces them.
His view was that if enough researchers reach common knowledge that they should all stop, they could break through the coordination barrier and stop together, instead of being replaced one by one by slightly less cautious people.