White House AI Adviser Says OpenAI, Anthropic Should Self-Limit Without New Rules

David Sacks's position places the administration in the unusual posture of endorsing voluntary AI slowdowns while simultaneously opposing the regulatory infrastructure that would enforce or verify...

David Sacks, who serves as chair of the President's Council of Advisors on Science and Technology (PCAST), said on September 14, 2026, that Anthropic and OpenAI should slow the development of their most advanced AI models if their own internal assessments conclude the technology poses serious risks, according to a Bloomberg interview published that day.

Sacks stated that no new federal regulation is necessary to achieve that outcome. His position, as described in the Bloomberg report, is that the companies themselves are capable of making the determination and acting on it without a statutory mandate or independent oversight body.

The PCAST chair also pushed back against proposals that would grant antitrust exemptions to frontier AI laboratories. Sacks argued, according to the Bloomberg report, that today's leading AI developers are themselves defining what the frontier looks like, a characterization that implies the companies hold both the competitive position and the safety-evaluation role simultaneously.

The Trump administration has pursued a deregulatory posture on artificial intelligence since taking office. In January 2025, President Trump signed an executive order revoking the Biden-era AI executive order from October 2023, which had directed federal agencies to develop safety standards and required developers of powerful AI systems to share safety test results with the government before public release, according to White House records.

The administration's approach relies on industry self-governance rather than agency rulemaking. No legislation creating a federal AI safety authority has passed Congress. The 119th Congress has seen multiple AI-related bills introduced in both chambers, but none has advanced to a floor vote in either the Senate or the House as of September 2026, according to the congressional record.

OpenAI and Anthropic each maintain internal safety teams and publish what they call model cards or system cards describing risk evaluations. Neither company is currently subject to a statutory requirement to disclose those evaluations to any federal agency, and no independent auditor is authorized by law to review their findings. What internal risk thresholds the companies use, and whether those thresholds have ever resulted in a delayed release, is not established in any public record reviewed for this report.

The antitrust dimension of Sacks's remarks carries regulatory weight. The Department of Justice and the Federal Trade Commission both have open investigative files related to the AI sector, though the scope and current status of those inquiries are not fully disclosed in public filings. Any antitrust exemption for AI developers would require an act of Congress. No such bill has been reported out of committee as of the date of this article, according to the congressional record.

Sacks's dual role as a federal advisory official and a figure with prior ties to the technology investment community has drawn ongoing scrutiny. His financial disclosures filed with the Office of Government Ethics are the authoritative record of any holdings or recusals relevant to his advisory work. Those filings are public documents; this report has not independently verified their current contents.

The policy debate over AI governance spans both parties. Some Republican members of Congress have called for lighter federal involvement, while some Democratic members have introduced bills requiring mandatory safety testing and federal certification before deployment of high-capability models. Neither camp has assembled the votes to move legislation through both chambers.

What remains unknown is whether Sacks's remarks reflect an official PCAST recommendation, a formal White House policy position, or his personal view. PCAST recommendations are transmitted to the president in formal reports; no such report specific to AI slowdown criteria has been published on the PCAST website as of September 14, 2026. A published PCAST report or a White House policy directive would be the public records that would resolve that question.