1 hour ago
34
A researcher who spent four months at Anthropic before quitting in protest has become one of the most-discussed figures in AI safety circles, and he wants to be clear about one thing: he did it alone.
Jacob Coxon, a 27-year-old British AI researcher, resigned from Anthropic on September 8, 2026, publicly warning that the industry’s race toward self-improving superintelligent AI poses catastrophic risks. Ahead of his departure, he consulted counsel from a nonprofit and briefed a journalist, but he denies that any third party coordinated or orchestrated his resignation.
Why a four-month tenure made this much noise
Coxon’s warnings spread fast. His post on X describing his reasons for leaving drew over 100 million views, an unusual level of traction for an essay-length thread about AI alignment written by someone most people had never heard of.
Coxon joined Anthropic in May 2026 after previously working at OpenAI from 2023 through early 2026, where he contributed to GPT-4o. The recruitment process was lengthy, which makes his exit timeline striking: he left without waiting for the standard six-month equity vesting threshold, effectively walking away from meaningful compensation to make a point.
His core concern centers on the trajectory toward artificial superintelligence, specifically systems capable of recursive self-improvement. He argues that the competitive dynamics between leading AI labs make it structurally difficult for any single company to slow down, even if its own researchers believe the pace is unsafe.
When the alignment lead agrees with the person who just quit
Evan Hubinger, Anthropic’s alignment science lead, publicly acknowledged that Coxon’s fears are legitimate. More specifically, Hubinger estimated a greater than 10% probability that AI could cause human extinction within the next decade.
Hubinger also admitted that Anthropic does not currently have a clear strategy for aligning a superintelligent system with human-friendly values.
Anthropic was founded in 2021 partly by former OpenAI researchers who believed the parent company was moving too fast on safety. The company has consistently positioned itself as the safety-conscious alternative in the frontier AI race. Coxon’s departure and Hubinger’s response complicate that framing, not because safety is not a priority at Anthropic, but because even prioritizing it does not appear to resolve the underlying problem.
What this means for the broader AI landscape
Coxon’s decision to consult a nonprofit’s legal counsel before resigning and to brief a journalist in advance suggests a deliberate, if individually driven, effort to document and publicize his concerns rather than simply move on to another job.
The denial of third-party collaboration matters in this context because critics of public AI dissent have frequently characterized such departures as coordinated advocacy campaigns rather than genuine individual conscience. Coxon’s account pushes back on that framing while acknowledging the practical steps he took to make his exit legible to a public audience.
Anthropic has attracted significant capital from backers who cite its safety focus as a differentiating factor. A scenario where the company’s own alignment lead assigns double-digit probability to extinction-level outcomes, while simultaneously admitting the absence of a concrete alignment roadmap for superintelligence, is not the kind of disclosure that fits neatly into a standard risk section.
AI development in the US is increasingly framed against the backdrop of Chinese AI advancement, and the argument that safety measures could create asymmetric disadvantages has been used to resist slowdowns. Researchers like Coxon are essentially arguing the opposite: that winning a race to build something no one knows how to control is not winning in any meaningful sense.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·