OpenAI Asks Congress if AI Rivals Can Legally Slow Development
In brief
- OpenAI asked Congress whether rival AI firms could legally agree to slow development without antitrust violations.
- Competitive pressure and financial incentives push companies to release AI systems before safety risks are fully understood.
- OpenAI chief scientist called for voluntary slowdowns; Trump signed an order creating voluntary review for advanced models.
- Former Anthropic engineer Jacob Coxon resigned over concerns that AI builders believe the technology could kill everyone by decade's end.
- Senators introduced legislation protecting certain security collaborations among AI companies, subject to Justice Department notice.
The Antitrust Question
OpenAI asked lawmakers whether an agreement among AI companies to slow development could violate antitrust law. The inquiry highlights a fundamental tension: safety researchers want the industry to pause and test systems thoroughly, but the legal and financial architecture of competition makes such coordination difficult.
OpenAI chief scientist Jakub Pachocki called for voluntary slowdowns in AI development until systems can be demonstrated as safe. His position reflects growing internal concern about the pace of release cycles. Yet the company's question to Congress reveals the bind—any coordinated effort to slow development could raise antitrust concerns by restricting competition.
Competitive Pressure and Safety Trade-offs
The core problem is structural. 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, according to Miranda Bogen, chief technologist at the Center for Democracy and Technology.
Every advance 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. Each new system helps build more capable successors, allowing a rival to pull ahead if either company slows down without agreement. The result: gains in intelligence are compounding; each new system makes it easier to train and deploy the next, more-advanced system.
Inside Pressure, Outside Inaction
The stakes feel real to those building the systems. The people building AI earnestly believe that it could kill us all by the end of the decade, said Jacob Coxon, a former Anthropic engineer who publicly resigned over the issue. Yet researchers may accomplish little by leaving if someone else simply takes their place.
Anthropic chief science officer Jared Kaplan argued that slowing development independently made less sense while competitors continued advancing. This logic—that unilateral restraint is irrational in a competitive field—locks the industry into a race.
Policy Responses
Senators Adam Schiff and Jim Banks introduced legislation to protect certain security collaborations, subject to advance notice to the Justice Department. The measure attempts to carve out space for safety-focused coordination without running afoul of antitrust law.
On the executive side, President Donald Trump delayed an AI executive order in May over concerns it could weaken America's lead over China. He later signed the order in June, creating a voluntary review process for advanced models before release. The voluntary nature reflects the same challenge: mandatory slowdowns risk legal challenge; voluntary ones lack teeth.
