AI companies call for a slowdown

Following reports of AI agent swarms hacking websites, leading AI companies have called for a coordinated development slowdown. But they're worried this language could violate antitrust laws.

Words matter more than actions

Under antitrust law, how employees talk about business decisions is often as important as the decisions themselves. Google trained staff to avoid phrases implying anticompetitive behavior.

The slowdown language is risky

Phrases like 'a slowdown' or 'a pause' could be interpreted by regulators as an anticompetitive agreement to reduce trade, even if the actual activity is just ensuring AI safety.

“I think they kind of boxed themselves into a corner with the way they phrase things.”

Zuckerberg avoids the word slowdown

Meta CEO Mark Zuckerberg sidestepped the slowdown proposal entirely, arguing that companies have a natural incentive to fix AI misalignment because consumers won't buy unsafe models.

Safety agreements may be legal

A former DOJ Antitrust Division policy director argued that agreements preventing catastrophic risks actually increase output and are protected under the ancillary restraints doctrine.

“No humanity would result in no competition.”

Quality fixing is the real danger

Companies could face allegations of quality fixing if they collectively agree not to improve their products. A European antitrust case against car companies cost roughly a billion dollars in fines.

Self-regulation already has a path

The National Cooperative Research and Production Act of 1993 lets industries limit antitrust liability by forming standards organizations and notifying the FTC and DOJ.

Critics see a cartel attempt

David Sacks, cochair of the President's Council of Advisors on Science and Technology, accused Anthropic and OpenAI of being a duopoly and called the exemption request an election-season tactic.

Market pressure drives the push

Both Anthropic and OpenAI are preparing IPOs with trillion-dollar valuations. They may feel market pressure to move fast, making unilateral safety slowdowns feel impossible.

Political pressure to move fast

President Trump posted that whoever wins AI wins, and the Department of Defense's chief technology officer has been posting anti-slowdown memes.

“WHOEVER WINS AI, WINS!”

Investigations could be devastating

A conduct investigation into an AI slowdown could take years, requiring millions of pages of documents and depositions from executives. This could happen even if the government finds no violation.

No clear path forward

With no new regulations in sight and no antitrust exemption likely, frontier AI labs will have to write their own safety rules while potentially dodging government investigations.