A stark statement from within the artificial intelligence industry has revived the debate over the real risks of a technology advancing at a rapid pace. Evan Hubinger, head of AI alignment and safety research at Anthropic, said that, in his personal assessment, there is more than a 10% chance that artificial intelligence could cause the extinction of humanity within the next 10 years.
The warning comes from a key voice in the field
Hubinger made the comment on X, where he wrote that he sincerely believes artificial intelligence could kill all humans. He stressed that his personal estimate for this scenario is above 10% within the next decade.
The statement carries weight not only because of its alarming content, but also because it comes from a researcher directly engaged with questions of AI safety and the alignment of AI with human interests.
Open admission of a lack of a solution
In the same remarks, Hubinger acknowledged that despite Anthropic’s work, the company still does not have a plan to solve the problem of aligning superintelligence with human values and interests.
Even more troubling, according to him, is that it is not clear whether real progress is being made toward a solution. This places the debate over AI safety before a sharp contradiction: development is continuing at speed, while guarantees still appear unclear.
The resignation that added pressure on the industry
Hubinger’s statement came after the resignation of researcher Jacob Coxon, who had worked at both OpenAI and Anthropic.
Coxon accused both companies of racing to create a self-improving superintelligence, warning that this approach is gambling with human lives. Such accusations add to doubts about how major AI companies are managing the technological race and the boundaries of safety.
The risk, he says, is not tied to current models
Hubinger clarified that his concern is not directly related to the models in use today, which he said remain low-risk.
His alarm concerns future systems that may be able to improve themselves and move beyond human control. That scenario is at the center of the global debate over AI, where promises of technological progress are advancing in parallel with fears of consequences that may later become impossible to manage.
Personal assessment, not the company’s official position
According to reports by Forbes and The Wall Street Journal, the figure of more than 10% does not represent an official Anthropic forecast, but rather Hubinger’s personal assessment.
Even so, the fact that such a warning comes from a senior figure in AI safety research is expected to further intensify pressure for transparency and accountability in an industry seeking to build ever more powerful systems.
The statement has brought back to the forefront a question that technology companies often treat amid promises of innovation: how controllable is the development of superintelligence in reality?
Beyond the alarming tone, the message emerging from this debate is clear: even within the industry itself, there is no complete certainty about the limits and consequences of the next phase of artificial intelligence.