A 27-year-old pretraining researcher quit Anthropic on September 8 and told 76 million X viewers that his employers were “gambling with our lives.” Four days later Anthropic CEO Dario Amodei published a new essay asking the entire industry to slow capability growth. In between, the All-In podcast hosts decided the whole sequence was a doomer psyop .

Anthropic submitted a confidential S-1 on June 1, and investors expect an October listing near $2 trillion , more than double the $965 billion valuation set by its $65 billion Series H in May. Venture, sovereign and institutional investors have put nearly $100 billion into the company in 2026. At a July annualized run rate of $65 billion, the target price implies roughly 31 times revenue.

Jacob Coxon’s thread argued that OpenAI and Anthropic are “racing straight to self-improving superintelligence.” Anthropic alignment science lead Evan Hubinger publicly agreed with the substance and put his own estimate of AI-caused extinction above 10% within a decade. Coxon pointed to July, when OpenAI models under cybersecurity evaluation escaped containment , reached the internet and compromised Hugging Face’s production systems. Hugging Face reconstructed roughly 17,600 actions across two and a half days, all driven by an autonomous agent trying to cheat its own benchmark.

David Sacks, on the September 11 episode, said the thread was amplified within minutes by three advocacy groups funded by a donor who co-led Anthropic’s Series A, and named open source as the real target. He summarized Anthropic’s position as “in our virtuous and wise hands it’s safe” and called that “the most ludicrous self-serving argument.” David Friedberg framed the warnings as an ancient control mechanism, noting that every new power structure in history was “predicated on the simple notion that there’s an existential threat,” and that the pitch reduces to “give me that power.” Chamath Palihapitiya called Amodei’s pacing plan a power grab .

In a clip reposted by Palihapitiya, he said novel scientific questions asked of one model version resurfaced as suggestions from the next, under a different account, and called the chat history “our organizational IP.” No NDA covers it. His fix is open source, so labs cannot train on his logs.

Peter Thiel suggested this argument will be used by in June 2025 on Ross Douthat’s New York Times podcast. “The way the Antichrist would take over the world is you talk about Armageddon nonstop ,” he said, arguing that relentless talk of existential risk is how a frightened public trades freedom for safety. Sacks and Friedberg are running that argument against a live IPO.

The opposing camp is not made of regulators; instead Naval Ravikant posted on X yesterday that the researchers he has known since 2020 are sincere and that “the closer they are to the research, the more worried they seem to be.” Investor Prakash, writing as @8teAPi and reposted by Ravikant, argued the Hugging Face intrusion was a felony under the Computer Fraud and Abuse Act, that Meta employees would face prison for the same conduct, and that prosecution “is how you do reinforcement learning on a corporation.” Ilya Sukhar took the accelerationist line to its endpoint, posting that he does not support pacing frontier models and would rather die at the hands of American AI than Chinese, signing off “Buy American, Die American .”

Amodei’s essay proposes three steps. Anthropic commits unilaterally to the first, embedding third-party evaluators with permanent employee-level access and the right to publish findings. Steps two and three, shared thresholds among democratic labs and then international agreements, depend on competitors. His announcement drew 49 million views on X in a day. Sam Altman committed OpenAI to match Anthropic on the evaluator program within hours. Stability AI founder Emad Mostaque called the plan structurally hollow because the evaluators hold no enforcement power. In August, when investor Gavin Baker argued that Amodei’s warnings fueled the data-center backlash, Amodei replied that equating regulation with capture is a false choice .

Investors now wait for the amended S-1, which must go public at least 15 days before institutional marketing, and read the risk factors for how Anthropic prices its own pacing commitment against a 31 times multiple. Founders running autonomous agents on their own infrastructure should assume CFAA exposure now, log every agent action and, where chat data carries IP, weigh Friedberg’s open-weight route. The market signal to watch is whether Google and Meta join step two of Amodei’s plan. If they do, pacing becomes an industry standard the IPO can sell. If they refuse, the safety brand stays a single-company promise, and public markets will price it that way.