Former Anthropic researcher leaves company over fears of 'out-of-control AI'

🛡 Former Anthropic researcher leaves company over fears of 'out-of-control AI'

On September 9, Jacob Coxon, who spent three years working on pretraining research at OpenAI and Anthropic, announced his departure from Anthropic on X. In his statement, he claims that 'neither company is acting responsibly': both, according to him, are 'racing toward a self-improving superintelligent system, risking lives.' The WSJ headlined the news as a departure over fears of 'out-of-control AI'. Coxon's profile on Google Scholar confirms his affiliation with Anthropic: his main publication is co-authorship on the GPT-4o system card (2024), and he also has a paper on weight-sparse transformers (2025).

🌍 The public departure of a pretraining researcher with accusations of a 'race to superintelligence' is a rare signal that debates about risks within the lab are going beyond official statements. The weight of the argument is limited, however: Coxon's h-index is 3, almost all citations come from the GPT-4o system card, and part of the community in the Hacker News thread (126 points, 137 comments) interprets the statement as attention marketing rather than a moral act.

👤 A specific example of how the discussion of AI existential risks translates into personnel decisions at labs: the statement on X and the thread on HN provide both sides of the argument — 'the race is real and dangerous' and 'this is PR, not a proven threat.'

Source 1: https://x.com/hilbertspaess/status/2097476196791709843 Source 2: https://www.wsj.com/tech/ai/anthropic-researcher-quits-over-out-of-control-ai-fears-707b7628