Antitrust lawyers warn AI companies against using the words “slowdown” and “pause” for safety initiatives
Legal experts warn that the words “slowdown” and “pause”, which AI companies use in connection with safety initiatives, may be interpreted under the Sherman Act as an anticompetitive agreement to restrict output, rather than as a legitimate safety measure.
Antitrust lawyers warn that the way AI companies talk about safety measures for their models can cause problems under antitrust law—specifically the Sherman Act, a key US law supporting a competitive market. According to experts, the problem lies not in slowing development itself to prevent risks, but in words such as “slowdown” or “pause”, which can give regulators the impression of an agreed restriction on output among competitors.
According to the article, Google previously trained employees to avoid phrases that could suggest anticompetitive conduct and instead emphasize benefits to consumers. John Bergmayer of the nonprofit organization Public Knowledge says that AI companies have “painted themselves into a corner” through the way they framed the issue, and that instead of talking about collectively slowing development, they should talk about cooperating on safety protocols against catastrophic risks. Mark Zuckerberg, CEO of Meta (his company recently escaped a major antitrust lawsuit from the FTC), did not explicitly endorse a formal “slowdown” and argued that labs have a natural incentive to improve model behavior because those that fail to do so will lose competitiveness.
Lawyers disagree on how risky such agreements actually are. David Lawrence, formerly policy director of the antitrust division at the US Department of Justice (DOJ), wrote that agreements preventing catastrophic risks actually increase output and support competition, and are protected by the so-called “ancillary restraints doctrine”. By contrast, Roger Alford, a law professor at Notre Dame and former second-highest-ranking official in the antitrust division at DOJ, warns that a joint agreement to refrain from introducing safety measures could expose companies to accusations of “quality fixing”—he refers to a European case involving automakers that agreed not to improve emissions technologies beyond legal requirements and ended up with a fine in the region of a billion dollars.
According to the article, David Sacks, co-chair of the presidential Council on Science and Technology, accused Anthropic and OpenAI of being a duopoly and called the request for an antitrust exemption a “pre-election psyop” and an attempt to create a cartel. Anthropic and OpenAI filed confidential IPO documents in the summer with valuations approaching or exceeding a trillion dollars; according to the article, Anthropic is expected to go public next month, while Sam Altman, CEO of OpenAI, said the company would postpone its IPO until 2027 because of safety concerns. The source text is cut off, and further details on this point are missing from the report—you can find the details in the source article.
Why it matters
For AI companies, this presents a concrete legal risk: public statements about a “slowdown” or “pause” in development for safety reasons can in themselves attract attention from the FTC or DOJ as a possible indication of a cartel agreement to restrict output, even though cooperation on safety itself may be legally protected. Word choice in communications about safety initiatives therefore has real consequences—imprecise wording can trigger an investigation that could be avoided by describing the same measure differently.
Relevant practical impact
What this means
For a business
AI companies coordinating safety measures risk phrases such as “slowdown” or “pause” being interpreted as an agreement to restrict output under the Sherman Act, which increases the risk of antitrust investigations regardless of the actual purpose of the measures.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.