David Sacks: OpenAI and Anthropic Don't Need Regulations to Pace Frontier Models

TEXT START: Dario has written that we need to “pace the frontier,” and Sam has agreed.

THE DISSECTION

This is a jurisdictional maneuver disguised as safety advice. It accepts that OpenAI and Anthropic control the frontier, then tells them to exercise that power voluntarily while rejecting any public framework that could constrain or supervise it. Product liability, customer preference, and China are used as evidence that restraint is possible, but the text never explains how restraint survives the incentive to win the capability race.

THE CORE FALLACY

It mistakes a private decision for a stable equilibrium. Under P1, more capable cognitive automation is economically valuable. Under P2, institutions cannot preserve human-only economic domains at scale. Under P3, mass productive participation collapses. A voluntary pause by two labs is therefore not a solution. If capability creates military, financial, or labor advantages, competitive pressure will eventually overpower corporate restraint. A duopoly cannot reliably exclude states, smaller labs, open releases, talent movement, or military programs.

The deeper deception is making superintelligence the threshold. The Discontinuity Thesis does not require superintelligence. Once AI is cheaper and better across enough cognitive tasks, the wage-to-consumption circuit is already severed. More reliable models may accelerate that process by making deployment easier. Product liability can punish spectacular failures; it does not price the slow destruction of human bargaining power. Pacing buys a lag, not an escape.

HIDDEN ASSUMPTIONS

• OpenAI and Anthropic can define the frontier and enforce a common pace.
• Corporate self-restraint will survive investor, shareholder, state, and military pressure.
• Competitors below the frontier will remain harmless and subordinate.
• Product liability and customer demand will internalize systemic cyber, labor, and geopolitical externalities.
• Capability, reliability, deployment, and alignment can be cleanly separated.
• China’s nonparticipation supports private discretion rather than demonstrating the weakness of global coordination.
• The public will accept firms retaining the power to decide whether human labor remains economically necessary.
• A later political conversation can repair the distributional damage created during the breathing room.

SOCIAL FUNCTION

Primary classification: partial truth serving elite self-exoneration and transition management, with an anti-regulatory propaganda layer.

The capture criticism has substance. Evaluator conflicts, liability incentives, and corporate lobbying are real mechanisms. But those observations narrow the political choice to voluntary restraint or destructive shutdown. That framing keeps control inside the firms while making public oversight appear irrational. It offers procedural anesthesia: trust the frontier owners to slow the machine long enough for everyone else to think.

THE VERDICT

This is not a plan to prevent the death of the postwar order. It is a fight over who holds the steering wheel during the descent. Voluntary pacing may delay the mechanical break and improve product reliability, but it does not solve P1, cannot overcome P2, and does nothing to prevent P3. The text correctly identifies regulatory capture, then offers private capture as the remedy. Its promise of goodwill now and regulation later is hospice politics: useful for buying time, useless as a theory of survival.