Jacob Coxon's exit from Anthropic tests a $2 trillion IPO pitched on safety

Daniel Kokotajlo, who resigned from OpenAI in 2024, recalls a researcher telling him that quitting a lab was akin to "renouncing your citizenship. Now you're a nobody. Who is going to protect you?" ^[16] The usual answer is the rival lab. According to New York Magazine, people who leave one lab often go straight to another ^[15]. Coxon took that route once. He spent about three years in pretraining roles at OpenAI ^[8], then moved to Anthropic earlier this year because of its reputation for model safety ^[5]. His second exit came less than nine months later ^[24]. This time he is leaving the industry ^[7], and he holds no ongoing financial stake in Anthropic ^[17].

Staff retention is where the case for a discount is weakest. To Vima's report calls Coxon's exit one of the first cases of an Anthropic employee leaving over safety fears. The only other one it cites from this year is a safety researcher who left to study poetry, warning that "the world is in peril" ^[9]. The same outlet describes Coxon as one of hundreds of researchers just like him across Silicon Valley's AI companies ^[18]. Two departures against that population do not make an attrition rate. The reports do not include headcount or turnover figures, so a retention discount on Anthropic cannot be sized.

The governance exposure an investor can price is at the top. Anthropic is seeking a $2 trillion valuation in an IPO expected to rank among the largest ever, and it has emphasized responsible development when courting investors ^[19]. Amodei and other company leaders have repeatedly urged the industry to slow development ^[14]. Amodei said he agreed with Coxon more than he disagreed with him ^[11]. Coxon's timeline is short. "We're on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already," he said ^[25]. He added that safety trade-offs are inevitable when labs compete with one another and with Chinese upstarts ^[6]. Anthropic did not immediately comment on the departure ^[22].

In one outcome, the slowdown becomes binding and coordinated, and growth slows with it. According to To Vima, AI boosters see Coxon's exit as part of a coordinated push for global regulations favorable to the biggest companies ^[20]. In another, the slowdown stays talk. There are no federal AI regulations, and the administration prefers a light touch ^[13]. President Trump said a "strong and smart" president is the only guardrail needed, and China dismissed the concerns as "fearmongering" ^[12]. In a third, more researchers follow Coxon out before the listing, and the retention problem the record does not yet show becomes one the prospectus has to disclose.

I think the second outcome is the likeliest in the near term, given Washington's posture. If it holds, Coxon's exit changes Anthropic's risk disclosures more than its business. The case against that view is the first outcome, where lab leaders who have already agreed to slow down turn the agreement into rules. The retention thesis would be proved right by further public exits from Anthropic's capabilities teams before the IPO, or by turnover figures in the filing.