Hyderabad: An Artificial Intelligence (AI) safety researcher at Anthropic has raised concerns over the rapid development of AI, saying he believes the technology has more than a 10 per cent chance of causing human extinction within the next decade.

What did the researcher say?

Evan Hubinger, who works on AI alignment at Anthropic, made the remarks on September 8 in response to a post by former Anthropic researcher Jacob Coxon on X.

Coxon announced his resignation from Anthropic and warned that AI companies were moving towards self-improving superintelligence without adequately addressing the risks associated with it.

Hubinger responded to Coxon, saying that Anthropic researchers “really do earnestly believe AI could kill all humans”. He added that he personally believed the probability was “>10% within the next decade”.

Hubinger has also discussed concerns around aligning future superintelligent AI systems with human interests.

What is the concern?

The comments are about potential future AI systems, particularly highly capable systems that could become capable of improving their own abilities.

The concern among AI safety researchers is that increasingly capable systems could become difficult for humans to control or could behave in ways that conflict with human interests.

Hubinger’s statement does not mean that current AI systems have a more than 10 per cent chance of killing humanity. The figure is his personal estimate of a future risk and is not an official probability issued by Anthropic.

Why did Jacob Coxon resign?

Coxon said he was leaving Anthropic because he believed leading AI companies were moving too quickly towards self-improving superintelligence.

In his posts announcing his resignation, Coxon said he had spent the previous three years doing pretraining research at OpenAI and Anthropic. He accused both companies of “racing straight to self-improving superintelligence” and “gambling with our lives”.

Coxon also argued that people developing advanced AI systems are aware of the possibility that AI could pose an existential threat, but that competition between companies could be encouraging them to continue developing increasingly powerful systems.

What does the 10 per cent figure mean?

The more than 10 per cent figure should not be treated as an established prediction that humanity will be wiped out.

It is Hubinger’s personal assessment of a potential future risk. Such figures are estimates based on researchers’ judgement and should not be treated as established probabilities or predictions.

The debate is part of wider concerns over the development of advanced AI systems and whether safety measures can keep pace with their capabilities.

For now, the statements by Hubinger and Coxon concern hypothetical future AI systems. They do not establish that current AI systems are capable of causing human extinction.

Elon Musk has also warned about AI risks

Elon Musk has previously expressed concerns about the possibility of AI causing human extinction.

In a 2025 interview on The Joe Rogan Experience, Musk said he believed there was an 80 per cent chance of a positive outcome from AI and a 20 per cent chance of “annihilation”. He also said he expected AI to become more intelligent than humans and eventually more intelligent than all humans combined.

Musk’s comments were not a response to Hubinger’s recent assessment. His 20 per cent figure was also not specifically an estimate of the probability of human extinction within a decade. Instead, Musk was giving his broader assessment of the possible outcomes of AI.

Other prominent AI researchers have also warned about the potential risks from advanced AI. Geoffrey Hinton, one of the pioneers of modern AI, has previously estimated a 10 per cent to 20 per cent chance that AI could cause human extinction within 30 years.