White House AI adviser David Sacks is challenging leading AI companies over their calls for stronger government oversight, arguing that laboratories such as Anthropic and OpenAI should slow the development of advanced systems themselves if they genuinely believe the technology poses serious risks.
Sacks’ criticism targets what he views as a contradiction in the industry’s approach: AI companies can voluntarily reduce the pace of frontier development, he argues, without demanding new regulatory frameworks, special antitrust protections or government-backed oversight mechanisms in return.
Sacks: If You Want to Slow AI, Do It Yourself
Sacks’ argument centers on a relatively simple proposition: companies developing increasingly capable AI systems control their own development decisions.
If Anthropic, OpenAI or other frontier laboratories believe that progress toward artificial superintelligence is moving too quickly, they could voluntarily reduce their development efforts rather than asking Washington to establish rules that apply across the industry.
In his criticism, Sacks argued that companies should “pace the frontier” themselves rather than relying on government intervention to impose their preferred restrictions.
His broader point is that voluntary restraint would provide a clearer demonstration that safety concerns are genuine.
Regulatory Demands Raise Questions About Motives
The dispute goes beyond the question of AI safety.
Sacks has argued that requiring the government to establish specific regulatory frameworks could produce unintended consequences for competition. Large AI companies possess considerably more financial resources and infrastructure than smaller startups, meaning expensive compliance requirements could potentially create barriers that smaller developers struggle to overcome.
Independent evaluations, licensing systems and other regulatory requirements could therefore have two effects at once: improving oversight while also making it harder for new competitors to enter the market.
That possibility has fueled accusations that some AI companies could benefit commercially from regulations presented primarily as safety measures.
Sacks’ criticism is essentially that companies should not condition voluntary restraint on Washington adopting rules that happen to strengthen the position of established AI laboratories.
“The Easiest Way” to Stop Superintelligence
Sacks also challenged the industry’s discussion around artificial superintelligence.
His argument was that if companies genuinely believe developing superintelligent systems would be dangerous, the most straightforward solution would be for those companies to agree not to build them.
That would eliminate the need to create elaborate regulatory mechanisms specifically designed to prevent development that the companies themselves could simply choose to avoid.
The challenge puts the responsibility directly on AI developers: demonstrate commitment to safety through corporate decisions rather than asking policymakers to impose restrictions on the entire industry.
Sacks Warns Against “Regulatory Theater”
The sharpest part of Sacks’ criticism concerns what he considers a mismatch between public safety rhetoric and regulatory demands.
He warned that asking government to provide a preferred regulatory structure in exchange for slowing AI development could appear less like genuine safety advocacy and more like political pressure.
In that interpretation, companies could use fears surrounding advanced AI to persuade policymakers to establish rules that protect incumbent laboratories from competition.
Sacks described that possibility as a form of regulatory capture, arguing that the public and political system should not have to accept industry-designed rules as the price for AI companies exercising restraint.
The Debate Is Bigger Than Anthropic and OpenAI
The disagreement reflects a much broader question facing the AI industry.
How should society regulate technologies that are developing faster than existing laws while being built primarily by private companies?
One approach is to establish government requirements covering development, testing and deployment.
Another is to allow companies to voluntarily establish safety limits.
A third approach would combine corporate responsibility with targeted government oversight focused on clearly defined risks.
The difficulty is determining where legitimate safety regulation ends and market protection begins.
For Sacks, the distinction is straightforward: if frontier AI companies believe a particular capability should not be developed, they have the ability to stop pursuing it without waiting for Congress or regulators to mandate the decision.
That leaves the industry facing a difficult credibility test. If AI laboratories continue pushing for stronger regulation while simultaneously advancing the frontier, critics are likely to keep questioning whether the primary objective is public safety, competitive protection—or some combination of both.
