Skip to content
English
FikirPilot content

According to an Anthropic researcher, the chance of AI “killing all humans” exceeds 10%

Updated: 10 Eyl 2026 · 2 min read · 314 words

Published: · Story reached us: · Processing time: 7 h 8 min

According to an Anthropic researcher, the chance of AI “killing all humans” exceeds 10%
A dark data center corridor

Anthropic researcher Evan Hubinger said he believes there is a greater than 10% chance that artificial intelligence could kill all humans within 10 years. In response to an X post by Jacob Coxon, who resigned from Anthropic, Hubinger said they genuinely believe this. Stating that Anthropic is doing everything it can, Hubinger said they do not yet have a plan or clear roadmap for solving the superintelligence alignment problem. Coxon, who worked as a researcher at OpenAI and Anthropic for the past three years, had resigned after arguing that the companies were not acting responsibly enough. Coxon claimed that the companies were engaged in an uncontrolled race to achieve self-improving superintelligence, that these systems could acquire power and resources through superhuman capabilities, and that executives were afraid behind closed doors.

Why it matters

These statements show that concerns about artificial intelligence risks are not limited to criticism from outside the companies but are also shared by the researchers developing the systems. Hubinger’s acknowledgment that there is no plan or clear roadmap for superintelligence alignment, while saying that Anthropic is making an effort, highlights the gap between the safety objective and the current level of preparedness. Coxon’s departure following his experience at OpenAI and Anthropic, accompanied by criticism over responsibility and the uncontrolled race, makes this debate not only a technical issue but also one of corporate governance and researcher responsibility. The claim that executives are afraid behind closed doors also shows that the relationship between the institutions’ public positions and internal concerns remains unclear. The key unanswered question is what concrete plan these companies will use to manage the risks posed by systems claimed to be capable of self-improvement and possessing superhuman capabilities.

Background

Anthropic is not a new name in the FikirPilot archive: in the past 90 days, we have published 9 news stories mentioning it; the latest is dated September 10, 2026.

Source: Independent Türkçe