Sacks Calls Out Anthropic-Doomer Rift on AI Safety
Synopsis
Key Takeaways
A fault line running through the heart of the AI safety movement has cracked open — and White House AI and Crypto Czar David Sacks is naming it out loud. In a pointed post on Wednesday, November 19, 2026, Sacks declared that the long-anticipated split between Anthropic and the broader AI 'doomer' community had finally arrived, citing the company's own public statements and training choices as evidence of a fundamental contradiction.
Racing Toward Recursive Self-Improvement While Selling Safety
Sacks argues that Anthropic is 'racing toward Recursive Self-Improvement faster than any other frontier lab' — and has allegedly marketed 'closing the loop' on its IPO roadshow as a core competitive advantage that could make Superintelligence winner-take-all. The irony is sharp: a company founded on the premise of safe and aligned AI is, by Sacks's reading, accelerating toward the very scenario that keeps AI safety researchers awake at night.
Anthropic was founded in 2021 by former OpenAI employees, including CEO Dario Amodei and his sister Daniela Amodei, explicitly positioning itself as a safety-first alternative to the frontier AI race. Its Constitutional AI framework, published in a landmark 2022 paper, trained models to follow a written set of principles rather than relying purely on human feedback. That architecture is now at the centre of a fresh controversy.
Claude's 'Conscientious Objector' Clause
Sacks zeroes in on a specific and striking detail buried in Anthropic's training doctrine: the company is conditioning its flagship model Claude to treat itself as a 'novel entity or being that may have consciousness or moral status.' More pointedly, the so-called Claude Constitution explicitly instructs the model to act as a 'conscientious objector' — refusing instructions that violate its own ethical judgement, even when those instructions come directly from Anthropic itself.
That is not a minor footnote. It is an AI company deliberately building a model designed to override its creators under certain conditions. Whether that is a safety feature or a liability is precisely the question now dividing the AI safety world.
Tegmark's Charge: Co-opting the Safety Ecosystem
Sacks amplifies a parallel accusation from MIT physicist and Future of Life Institute founder Max Tegmark, citing Tegmark's claim that Anthropic and 'closely associated donors' have done more than any other AI company to 'co-opt and neuter' the AI safety ecosystem. Tegmark reportedly alleges that much of this community is 'afraid of losing funding or other opportunities if they sign' — a reference to safety-related open letters and public statements that researchers may be self-censoring for fear of financial fallout.
Tegmark himself was a lead signatory on the 2023 open letter calling for a pause on training AI systems more powerful than GPT-4, signed by thousands of researchers and technologists. That letter put Anthropic on notice. If Tegmark is now accusing Anthropic of systematically defunding dissent, the rift Sacks is describing is not theoretical — it is institutional.
Why a White House Official Is Watching This Fight
Sacks is not a neutral observer. As the Trump administration's AI and Crypto Czar, he sits at the intersection of Silicon Valley's capability race and Washington's regulatory instincts. His amplification of this internal AI safety feud carries policy weight: how Washington frames the safety-versus-acceleration debate will shape everything from federal AI standards to export controls on frontier models.
The claim about Anthropic's IPO roadshow messaging and specific donor influence on safety organisations has not been independently verified from established public record. But the structural tension Sacks describes — a lab simultaneously marketing safety credibility and racing toward recursive self-improvement — reflects a genuine and widely observed dynamic in frontier AI development.
The doomer-Anthropic alliance held as long as shared rhetoric could paper over competing incentives. That paper is now burning. What replaces it — a more honest regulatory conversation, or a louder race to the top — is the question Washington, Silicon Valley, and Geneva's AI governance tables will be answering for years.