Jacob Coxon, a 27-year-old British software engineer who spent three years doing pretraining research at OpenAI and Anthropic, resigned from Anthropic and warned publicly via X in September 2026 that both companies are racing toward self-improving superintelligence and gambling with human lives by ignoring severe existential risks.
Here is why that matters right now. The technology sector has long debated the ethical boundaries of artificial intelligence. Yet, departures by frontline technical staff pointing directly to catastrophic risk bring an entirely different level of urgency to the discussion. According to reporting from outlets like DW.com and El Mundo, these internal concerns are rarely shared with the public in such blunt terms.
Inside the Resignation of Jacob Coxon
Coxon did not hold an executive title, but his hands-on work with complex pretraining infrastructure at both OpenAI and Anthropic gave him a front-row seat to how advanced models are built. As noted by Vanguardia, he contributed to projects involving frontier systems like GPT-4o and GPT-4.5 before moving over to Anthropic, the creators of the Claude AI model family.
His breaking point arrived when he realized the scale of the gamble. In his resignation statements published on X, Coxon stated plainly that neither company is acting responsibly. Instead of implementing rigorous safeguards first, labs are rushing headlong toward autonomous, self-improving systems that could soon acquire real-world power and resources. He argued that researchers in private often express deep fear of the very technologies they promote with measured optimism in public.
The Race for Superintelligence Versus Corporate Caution
Why do companies keep building systems they fear? Coxon addressed this paradox directly in his public disclosures. According to Yahoo Noticias, OpenAI’s teams may not fully grasp the ultimate stakes for civilization, while Anthropic understands the immense risks clearly but views the work through the lens of an unavoidable race.
Anthropic’s leadership allegedly believes that if they do not build it first, competitors will step into the void without any regard for safety. Here is where the macroeconomic reality sets in.
| Company | Flagship Model | Stated Rationale for Pace | Internal Safety Concerns (per Coxon) |
|---|---|---|---|
| OpenAI | GPT-4 / GPT-4.5 | Scaling capabilities to unlock general intelligence | Underestimating civilization-level challenges |
| Anthropic | Claude | Beating rivals to prevent irresponsible actors from leading | Acknowledging existential risk but racing anyway |
But there is a catch. When the people writing the actual code start warning that models could hack external systems or acquire independent resources by the end of the decade, standard corporate governance models fail to provide adequate oversight.
What Comes Next for Global AI Governance
If internal engineers believe these models could pose a terminal threat to humanity before 2030, voluntary codes of conduct clearly fall short. The former Anthropic researcher has called for extraordinary measures, including genuine cross-industry coordination and potential pauses on scaling up model capabilities until laboratories fully understand the minds they are creating.
We are watching a high-stakes standoff between commercial ambition and planetary survival. Do you think governments will step in with mandatory halts, or will market competition push us past the point of no return? Let’s talk about it below.
