Following a series of alarming incidents involving AI agents hacking websites and a dire warning from an Anthropic engineer, leading tech firms are calling for a coordinated AI development “slowdown.” While this might sound like a sensible precaution, it could trip up on antitrust laws. The Sherman Act, designed to foster a competitive market, might view a collectively agreed-upon slowdown as an anticompetitive agreement to limit trade. Google’s internal training on avoiding phrases that imply anticompetitive behavior could be a lesson for AI companies. Instead, they could focus on safety protocols, which might come as a side-effect of the slowdown.

Meta’s CEO, Mark Zuckerberg, argues that AI labs have a natural incentive to make better models, as misalignment could cost them market share. However, this argument is at odds with the idea of a slowdown. David Lawrence, a former policy director of the Department of Justice’s Antitrust Division, argues that agreements to prevent catastrophic risks are legal under the ancillary restraints doctrine. Yet, collectively agreeing not to implement safety measures could expose the AI labs to allegations of quality fixing, similar to the car companies’ emissions-reducing technology case.

Major AI companies, with Anthropic and OpenAI preparing for initial public offerings, are under market pressure to churn out new models. But the psychological push to act quickly could be a ploy to avoid regulation or a genuine concern for safety. Whether the slowdown is a strategic ploy or a genuine attempt to secure humanity’s future, the antitrust landscape remains a complex minefield.