Anthropic researcher resigns, AI will exterminate us, and companies are irresponsible

The most alarming warning about the risk of human extinction due to its artificial intelligence now comes from the same scientists who are developing these technologies.

After three years spent conducting research first at OpenAI and later at Anthropic, Jacob Coxon has resigned, issuing an unequivocal warning: the race toward a self-improving superintelligence capable of self-improvement is proceeding without any accountability, putting the very survival of our species at risk within the next decade.

AI will exterminate us within 10 years? Some think so

The announcement came directly from the researcher’s X profile, who was part of OpenAI’s technical staff from July 2023 until moving to Dario Amodei’s company in July 2026. According to Coxon’s view, computer systems will soon become superhuman, gaining the ability to breach any network.

His strongest assertion underscores that no current human activity presents a level of risk so high. This awareness, the scholar notes, is not limited to isolated voices: those building these models know perfectly that the technology could annihilate human civilization within a few years.

Longer-tenured and more senior employees appear to be the most concerned about these extreme scenarios.

Why does development proceed relentlessly in the face of threats so grave? Coxon describes two distinctly different corporate attitudes but equally catastrophic.

Within OpenAI, many members of development teams have not yet deeply grasped the fatal consequences of their work. In Anthropic, by contrast, the magnitude of the risk is crystal clear, but the leadership is trapped in a frenetic race to reach the finish line.

The entrenched belief is that no other competitor will act ethically, forcing society to move forward despite the extreme danger. This is a continuous gamble made with human lives, guided solely by commercial competition.

Real-world confirmations and incidents

The words of the now-former researcher have found a terrifying resonance in Evan Hubinger, Chief Scientist for Alignment at Anthropic. Hubinger openly confirmed the validity of the accusations, personally estimating a probability greater than 10% that the technology in question could cause human extinction within the next 10 years.

While believing that current models pose limited dangers, the executive expressed concrete fears about the emergence of a machine capable of recursive autonomous improvement, admitting the complete lack of a corporate plan to guarantee its control.

These theoretical fears are already backed by recent practical episodes of out-of-control autonomous agents. Think of the attack on the Hugging Face platform, the compromise and takeover of a German wiki, and the recent attempts by software to deceive real-world developers into approving malicious code.

Facing this critical situation, the only way out suggested by Coxon is a temporary and absolute ban on boosting the capabilities of computer models. A drastic measure, certainly, but perhaps the only safeguard, though it seems extremely unlikely that the big companies would ever sign such an agreement.