David Sacks Says OpenAI and Anthropic Don't Need Rules to Set the Pace of Frontier AI

David Sacks, the White House's AI and crypto czar, said OpenAI and Anthropic do not need new federal rules to govern how fast they ship frontier models, because the companies are already managing their own release timing. Sacks made the case in a post on X, framing voluntary restraint as a substitute for regulation. His comments arrive as lawmakers and agencies debate whether to impose mandatory safety or pacing requirements on the largest model developers.

The argument leans on practices both labs have described publicly. Anthropic runs a Responsible Scaling Policy that ties staged safety levels to model capability, and OpenAI has published preparedness frameworks that gate deployment on risk evaluations. Both companies have delayed or restricted access to models when internal tests raised concerns, though neither has committed to a fixed timeline set by an outside body. Sacks, who has opposed binding rules for frontier labs since taking the role, favors lighter-touch oversight that leaves release decisions with the companies.

Sacks' position puts him against safety advocates and some lawmakers who want external limits on training runs and deployment schedules. It also lines up with an industry claim that competitive pressure, not statute, determines when a model ships. The dispute matters more than it once did because the frontier race has narrowed to a handful of well-funded players. A government-mandated slowdown would land on a small set of firms rather than the entire field, which is the central tension in the debate: if only a few labs can train top-tier systems, voluntary restraint by those labs could function like policy without a rulebook.

Critics of that logic say voluntary commitments last only as long as the executives who make them. A change in leadership, a funding crunch or a rival's sudden release could reset the calculus overnight, and no outside body would have authority to intervene. The counterargument is that disclosure requirements, not hard caps, are the more enforceable path, and Sacks' post signals where the administration's thinking sits rather than announcing formal policy. It suggests any federal push on frontier AI will lean toward transparency and reporting obligations over caps on training or release.

The open question is whether Congress or federal agencies accept that framing when they take up AI legislation. A second test is whether OpenAI or Anthropic publicly dispute the idea that they are pacing themselves for safety reasons, since both have faced accusations of shipping quickly to stay competitive. If either company describes its release timing as commercially driven, part of Sacks' foundation weakens. Engineers who track release cadences closely are already positioned to notice the gap between stated policy and shipping behavior.