David Sacks urges OpenAI and Anthropic to slow frontier AI voluntarily

Summary

David Sacks responds to calls by Anthropic CEO Dario Amodei to “pace the frontier,” saying OpenAI and Anthropic can slow development on their own. He characterizes the two companies as a duopoly in frontier intelligence by market share, revenue growth, and model capability, and says they have claimed their lead is widening through recursive self-improvement. Sacks says he would support a responsible decision to hold back unreleased models if the companies consider them dangerous, but argues they do not need anyone else’s permission to do so. He objects to proposals he portrays as requiring suspension of antitrust law, a cartel-like arrangement, or a regulatory approval process that supersedes product liability. He also challenges the independence of METR, alleging ties to Anthropic investors and staff, and questions relying on the same evaluators to oversee competitors that are not at the frontier. In his account, restraint may also serve business interests: companies face product-liability exposure if their products enable damaging cyberattacks, while customers value predictable behavior. Referring to a “Hugging Face episode” without detailing it, he says trading raw capability for reliability could be good business as well as alignment. Sacks argues that a voluntary slowdown could create room for a more considered regulatory debate, while noting his view that China is unlikely to join a global agreement and criticizing Bernie Sanders’s “shut it all down” position. He says the simplest way not to build superintelligence is for the labs to agree not to build it. But he warns that demanding a preferred regulatory framework in exchange could look like blackmail of the public and political system. He says voluntary restraint could build goodwill; otherwise, he suggests, the calls may be regulatory capture or an election-season “psyop.” These are Sacks’s claims and judgments, not independently established findings in the post.