David Sacks: Anthropic Suffers from Corporate Schizophrenia
David Sacks accused Anthropic of suffering from "corporate schizophrenia," stating that while it continuously issues apocalyptic warnings about artificial intelligence, it paradoxically rushes to develop cutting-edge models and seeks favorable regulatory frameworks.
Sacks, who is currently the co-chair of the President's Council of Advisors on Science and Technology and previously served as an AI and crypto advisor at the White House, has publicly attacked Anthropic for months: framing the safety narrative as regulatory capture, linking model releases with worst-case scenario studies as fear marketing, and stating "do not set up a DMV for AI."
Contradictory statements include: Dario Amodei previously claimed that half of entry-level knowledge jobs could disappear within one to five years, while Sacks pointed out that net employment remained stable a year later and that Anthropic was still hiring engineers with salaries around $570,000; Amodei called for "controlling the pace of frontier development," to which Sacks responded, "you are the frontier; the simplest way is for you to agree not to create superintelligence," framing accompanying regulatory demands as extortion against the public and political system.
At the same time, Anthropic researcher Jacob Coxon resigned and warned of potential extinction within a decade, while alignment science lead Evan Hubinger publicly stated a greater than 10% extinction probability within ten years; Amodei later published a call to slow down and introduce on-site evaluations. Sacks characterized third-party assessments like METR as entangled with Anthropic's investors and former employees, denying their independence.
Sacks' consistent stance advocates for open weights, multi-vendor competition, and prioritizing competition with China. He linked apocalyptic rhetoric, state-level legislative lobbying, and dual oligopoly market shares into the same accusation: first comparing products to nuclear bombs, then continuing to release models, and subsequently demanding the government set regulations according to their safety standards. Anthropic, on the other hand, continues to publish research on alignment, deception, and agency misconduct, using the narrative of responsibility as a differentiating selling point for corporate clients.
In market mechanisms, buyers are government and large enterprises seeking safety compliance, while sellers are laboratories simultaneously selling model capabilities and apocalyptic papers; funding flows from corporate contracts and potential IPO expectations to safety brands, while regulatory bandwidth shifts to "who has the right to hit the brakes." Beneficiaries are laboratories in power that can turn fear into entry barriers, while those under pressure are the open weights camp and White House technology advisors who do not want a new licensing system. The driving events stem from resignation posts and calls for slowing down, rather than new benchmark tests.
There is no exchange-verifiable "schizophrenia index"; what can be verified is the timeline of the same company issuing extinction probabilities, releasing new models, and demanding regulatory support.
Source: Public Information
ABAB AI Insight
Sacks rewrites Anthropic's safety discourse from a moral claim to a competitive strategy. The core structure remains unchanged: first warning of extinction and unemployment, then releasing stronger models, and finally demanding the government set regulations according to their assessment system. Corporate schizophrenia is not a medical diagnosis; it is his framing of the "safety super-ego" and "market share id" as the same balance sheet.
The capital pathway is that safety research becomes a procurement pass. Corporate clients want brand insurance, Washington requires citable risk assessments, and laboratories exchange papers for regulatory discourse power. METR's independence is attacked because once third-party assessments are deemed overlapping with investors, they can no longer grant social licenses to the dual oligopoly. Sacks represents the open weights alliance of Nvidia and Meta: multiple models, fewer gates, and no stopping in competition with China.
Similar structures appear in tobacco companies sponsoring health research, energy companies lobbying for both carbon taxes and expansion, and military contractors writing nuclear risk white papers. The industry's position is in a licensing competition during an expansion phase: whoever defines danger defines who can continue training. If the slowing initiative is tied to antitrust exemptions, it shifts from safety to market division.
Structural judgments belong to regulatory changes. The mechanism is that fear is used as a rent-seeking tool: the louder the warnings, the more exclusive assessments and access are needed; the more exclusive the assessments, the harder it is for newcomers to enter. The contradictions are not mistakes; they are two sides of the same strategy—without apocalypse, there are no gates; without gates, there are no dual oligopoly rents.
ABAB News · Law of Cognition
Warning of apocalypse while releasing new models means the gate is right at their doorstep.
Safety papers can be sold to clients and also to legislators.
The simplest brake is to stop oneself first, rather than legislate to stop others.