Sacks to Anthropic, OpenAI: don't wait for permission to slow AI

Sacks' post is a direct reply to a specific pair of statements: Dario Amodei's September 12 essay "We Must Pace the Frontier" and Sam Altman's post backing it. Sacks does not dispute that frontier labs may want to slow down. His objection is to the sequencing — he argues the slowdown itself is the easy part, and that framing it as requiring a regulatory framework inverts who is asking whom.

Much of his argument leans on the two companies' own claims. He notes that by market share, revenue growth and model capability, Anthropic and OpenAI hold an oligopoly on frontier AI, and that they themselves say recursive self-improvement is widening their lead. If that is true, he writes, then they — not regulators — are the ones setting the frontier, so the simplest way not to build superintelligence is for the two of them to agree not to. He also flags that China is very unlikely to join any global agreement, a factor he says must be weighed.

The backdrop gives the exchange its edge: Sacks is a former PayPal COO, Yammer founder and Craft Ventures co-founder who became the White House special adviser for AI and crypto, left that role in March 2026, and now co-chairs PCAST at OSTP alongside Nvidia's Jensen Huang, Meta's Mark Zuckerberg and AMD's Lisa Su. With US midterms in November, his closing line — that acting first earns credibility, while not acting reveals this as regulatory capture or election-season lobbying — frames the whole debate as a test of whether the labs' stated caution converts into action, and Sacks is betting the answer will be visible either way.