OpenAI has been working with Anthropic and Alphabet Inc’s (NASDAQ: GOOG) Google DeepMind on artificial intelligence safety as industry leaders call for slower development.
OpenAI global policy chief Chris Lehane revealed the discussions during a Washington press briefing on Tuesday. He said the three leading AI developers have held talks for several weeks.
Additionally, the talks come amid growing warnings from AI executives about rapidly improving systems. Anthropic CEO Dario Amodei called for slowing development of advanced AI models in an essay published over the weekend.
OpenAI CEO Sam Altman subsequently supported Amodei’s proposal and said his company had discussed similar concerns internally.
Amodei argued that AI capabilities have recently started advancing much faster. He attributed part of that acceleration to AI systems becoming increasingly capable of helping develop their successors.
However, coordinating a slowdown across competing companies could create a separate legal problem.
U.S. antitrust law generally prohibits competitors from coordinating to restrict market output. Some legal experts argue that an agreement to slow AI development could qualify as such a restriction.
Consequently, Amodei has called for the federal government to help enable discussions between competing AI developers. He proposed government mediation, a standards organization or a narrow antitrust waiver for certain safety discussions.
Congress has already considered legislation addressing the issue.
In July, Sens. Jim Banks and Adam Schiff introduced the Collaboration on Adversarial Threats and Security Risks Act. The bipartisan proposal would provide limited antitrust protections for companies collaborating on AI safety issues.
Meanwhile, some smaller AI companies have questioned whether such protections could entrench the industry’s largest developers.
Read more: Snowflake jumps up 22% on latest earnings and boosted guidance
Read more: OpenAI’s public launch gets put on hold over safety concerns
Federal regulators have also expressed skepticism about exemptions
Cohere CEO Aidan Gomez has argued that dominant AI companies could use safety rules to shape the market. Under that scenario, major developers could help establish standards that smaller competitors must follow.
“Once again using fear under the pretext of protecting the public, these oligopolies are now requesting to bend competition rules and be permitted to dictate the terms for everyone else,” Gomez wrote in an essay. “A wolf in sheep’s clothing, a cartel by another name.”
Federal regulators have also expressed skepticism about granting exemptions.
Federal Trade Commission chairman Andrew Ferguson said Tuesday that requests combining new regulations with antitrust exemptions made him deeply suspicious.
Ferguson said such arrangements could create barriers that protect established companies from new competitors. He made the remarks during an event at Georgetown University.
Additionally, Attorney General Todd Blanche declined to commit the Trump administration to granting an exemption.
Blanche told Bloomberg that he would consult other administration officials before determining how the government should approach the issue. President Donald Trump has generally opposed expanding federal AI regulation.
However, Lehane said OpenAI does not expect the companies to require an antitrust waiver.
He pointed to existing arrangements that allow competitors to exchange sensitive information when addressing safety concerns. Airlines, for example, can share certain information for safety purposes.
Furthermore, federal guidance has previously allowed companies to share information when defending against cybersecurity threats.
Some legal experts argue that AI safety coordination could fit within those existing protections. That could eliminate the need for Congress to create a new antitrust exemption.
The debate arrives as OpenAI increases its support for federal AI safety legislation.
Read more: Arm CEO sees AI accelerating cancer diagnosis, research and treatment
Read more: OpenAI Navier-Stokes breakthrough sparks data privacy controversy
Industry divided over government involvement
Lehane visited Washington this week to meet lawmakers and discuss potential safeguards. OpenAI has backed two bipartisan legislative efforts addressing risks from advanced AI systems.
One proposal is the FRONTIER Act, which would establish a federal framework for evaluating advanced AI.
The legislation would require the largest AI developers to provide access to independent verification organizations. Those organizations could examine whether companies follow their stated safety practices.
In addition, Amodei has proposed embedding independent evaluators directly inside frontier AI companies.
Anthropic has committed to adopting that approach voluntarily. Altman has also said OpenAI intends to follow Anthropic’s lead.
A second proposal remains under development in the Senate.
Senate Majority Leader John Thune is working with Republican Sen. Ted Cruz and Democratic Sen. Amy Klobuchar on the measure. The proposal would reportedly give the Commerce Secretary authority to deploy government auditors to test advanced AI models.
Meanwhile, the industry remains divided over how much government involvement those measures should require.
The Trump administration has generally favoured rapid AI development and lighter federal regulation. Administration officials have also argued that excessive restrictions could weaken American companies against Chinese competitors.
Amodei has acknowledged that geopolitical competition complicates efforts to slow development.
However, he argues that stronger export controls could preserve America’s technological advantage while companies devote more time to AI safety.
His broader proposal eventually calls for international coordination, including limited cooperation with China.
Such cooperation could focus on particularly dangerous applications, including AI-assisted development of biological weapons.
For now, the immediate question remains whether Washington will create a formal framework for domestic cooperation.
Thune said this week that Congress may not complete AI legislation before the end of the year.
