Sacks to Anthropic and OpenAI: Then Just Do It
Dario Amodei asked the industry to pace the frontier. David Sacks answered on September 13 with the one response nobody had scripted a rebuttal for: go ahead. The post is at https://x.com/DavidSacks/status/2098973625252708460 and it did roughly 57,000 likes, 9,500 reposts and 2,700 replies, plus 203 points on Hacker News. Sacks ran the White House AI and crypto portfolio from December 2024 until his 130 special-government-employee days ran out in March, and now chairs the President's Council of Advisors on Science and Technology, so this is not a random VC subtweet.
The argument is a trap dressed as an endorsement. "You guys are the frontier," he writes. "By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence." If the unreleased models are genuinely scary, slow down, he supports it. "But stop pretending you need anyone else's permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability." The move is to grant the safety premise entirely and then strip out the policy ask that was riding on it, which leaves the labs holding a decision they can execute unilaterally and no longer have a reason to trade for.
Then he goes after the one concrete commitment in Amodei's post. "Stop pretending METR is independent when it is intertwined with Anthropic's investors and staff." [The embedded-evaluator offer](https://clauday.com/article/7d70e566-1ce0-4e7e-b01d-2cc09b94ecfe) was the only falsifiable thing on the table, the piece of the proposal Anthropic could do by itself, and Sacks spent a sentence on it to make the point that a third-party auditor funded and staffed out of your own orbit is not a third party. Whether that lands depends on facts about METR's cap table that neither man laid out, but as a rhetorical matter it removes the load-bearing beam.
His alternative account of the motive is the part worth arguing with. Labs face real product-liability exposure if an agent enables a genuinely damaging cyberattack. Customers punish models that behave unpredictably. "After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want." That is a deflationary reading of the entire safety turn, and it is annoyingly hard to falsify, because the commercial and the sincere explanation predict the same behavior. He also concedes what the safety side usually has to fight for — China is unlikely to join any global agreement — while pointing out that this cuts against the pacing case, not for it.
The closer is the threat. "Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it. If you do, you'll buy goodwill for the next conversation. If you don't, we'll know this was just another bid for regulatory capture — or an election-season psyop." In the same 24 hours Xe Iaso published a satire at https://xeiaso.net/notes/2026/everyone-slowdown-but-me/ in which a fictional lab calls for a global pause that conveniently exempts itself, and it outscored every other AI item on Hacker News at 729 points. Two very different registers, one message: the proposal has been read as a business strategy, and Anthropic now has to prove otherwise by doing something instead of asking for something. The falsification test is cheap and public — name a capability line Anthropic will not cross, and a date.
← Back to all articles
The argument is a trap dressed as an endorsement. "You guys are the frontier," he writes. "By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence." If the unreleased models are genuinely scary, slow down, he supports it. "But stop pretending you need anyone else's permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability." The move is to grant the safety premise entirely and then strip out the policy ask that was riding on it, which leaves the labs holding a decision they can execute unilaterally and no longer have a reason to trade for.
Then he goes after the one concrete commitment in Amodei's post. "Stop pretending METR is independent when it is intertwined with Anthropic's investors and staff." [The embedded-evaluator offer](https://clauday.com/article/7d70e566-1ce0-4e7e-b01d-2cc09b94ecfe) was the only falsifiable thing on the table, the piece of the proposal Anthropic could do by itself, and Sacks spent a sentence on it to make the point that a third-party auditor funded and staffed out of your own orbit is not a third party. Whether that lands depends on facts about METR's cap table that neither man laid out, but as a rhetorical matter it removes the load-bearing beam.
His alternative account of the motive is the part worth arguing with. Labs face real product-liability exposure if an agent enables a genuinely damaging cyberattack. Customers punish models that behave unpredictably. "After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want." That is a deflationary reading of the entire safety turn, and it is annoyingly hard to falsify, because the commercial and the sincere explanation predict the same behavior. He also concedes what the safety side usually has to fight for — China is unlikely to join any global agreement — while pointing out that this cuts against the pacing case, not for it.
The closer is the threat. "Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it. If you do, you'll buy goodwill for the next conversation. If you don't, we'll know this was just another bid for regulatory capture — or an election-season psyop." In the same 24 hours Xe Iaso published a satire at https://xeiaso.net/notes/2026/everyone-slowdown-but-me/ in which a fictional lab calls for a global pause that conveniently exempts itself, and it outscored every other AI item on Hacker News at 729 points. Two very different registers, one message: the proposal has been read as a business strategy, and Anthropic now has to prove otherwise by doing something instead of asking for something. The falsification test is cheap and public — name a capability line Anthropic will not cross, and a date.
Comments