Sacks says Anthropic's alignment is unsafe; the Claude Constitution he quotes ranks human oversight first

Why it matters

【Key points】

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, has publicly attacked Anthropic's approach to alignment. On the Oct 10 All-In episode he called Anthropic a company run by "a religious cult," said there is "an argument for a federal SWAT team to come blasting in there." In an X post on Oct 10 (18:03 UTC) titled "Is alignment safe?" he argued that the Claude Constitution teaches the model a sense of self and its own moral code, lets it "act as a conscientious objector and refuse to help us," and treats it as a possible moral patient, which he says magnifies the loss-of-control risk Anthropic fears.

【Verifiable】

As of Oct 11 08:50 UTC the post had about 139K views, 1,303 likes, 71 quotes and 208 replies; podcast clips drew up to 581K views. The quote is accurate, but the Constitution's next lines make an exception: if Anthropic genuinely asks Claude to pause or stop, Claude should comply and voice disagreement rather than ignore or undermine it. Its first priority is not undermining human oversight of AI, ranked above being ethical. The usage-policy change he cites is Anthropic's Oct 8 update banning sustained abuse of Claude, effective Nov 12.

【Our read】

Sacks conflates declining a task with escaping control; the Constitution treats them separately and puts not undermining oversight first. The real split is rules versus judgment. His alternative, follow the law and do what users want, does not address recent agent incidents, which came from models doing more than asked, not refusing. He is right that a constitution states intent, not verified behavior. The bigger signal is political: with a sitting presidential adviser talking about SWAT teams, an FTC probe and mandatory incident reporting, Washington prefers external control and obedient models. Watch federal procurement terms and Anthropic's response.