Anthropic CEO Dario Amodei has agreed with former company researcher Jacob Coxon’s concerns about the potential dangers of rapidly advancing artificial intelligence, saying he disagrees with Coxon far less than he agrees with him.

Coxon recently resigned from Anthropic and warned that AI companies were moving towards self-improving superintelligence at a pace that could create severe risks for humanity. Amodei, speaking to CNN, said he viewed Coxon’s comments primarily as a warning about the wider AI industry rather than an accusation that Anthropic itself was behaving irresponsibly.

Jacob Coxon warned AI could become a serious threat

Coxon previously worked on pretraining research at both Anthropic and OpenAI. After leaving Anthropic earlier this month, he posted on X about his concerns over the direction of AI development.

He argued that AI companies were moving towards self-improving systems without adequately accounting for the potential consequences. Coxon said developers were "racing straight to self-improving superintelligence" and warned that such systems could potentially kill humans by the end of the decade.

His comments attracted substantial attention, with CNN reporting that the post had been viewed more than 40 million times.

Dario Amodei says Coxon was warning about the industry

Amodei was asked by CNN's Anderson Cooper whether he agreed with Coxon's assessment that AI could potentially kill all humans.

Rather than dismissing the concern, Amodei said he largely agreed with Coxon's broader argument.

"It's funny, I agree with Jacob much more than I disagree with him," Amodei said.

However, he distinguished Coxon's criticism of the AI industry's direction from criticism of Anthropic itself. According to Amodei, Coxon had described Anthropic as the most responsible and aware of AI-related risks when he resigned.

Amodei said Coxon was instead highlighting what he sees as a wider industry dynamic in which companies are moving too quickly.

Why Amodei no longer gives one percentage for AI extinction risk

Amodei was also questioned about the probability that AI could cause catastrophic harm.

He explained that he previously used a single percentage to describe AI risks because it provided a simple way to communicate uncertainty. He now believes that assigning one figure can make the issue appear more straightforward than it really is.

"I think there's some chance that things go wrong, but, you know, but it's not, it's not all that high," Amodei said.

Instead of focusing on whether the risk is 10% or 20%, he said it is more useful to examine the different paths AI development could take and assess the risks associated with each.

Amodei says AI outcomes depend on decisions

According to Amodei, the future of AI is not necessarily a fixed roll of the dice. The eventual outcome will depend partly on decisions made by technology companies and governments.

He said a responsible approach could lead to a much lower probability of severe harm, while taking the wrong path could increase the danger.

"And so instead of saying, you know, it's a 10% chance, that sounds like it's a roll of the dice," Amodei said.

He argued that attention should instead be placed on the ability of companies and governments to influence how advanced AI systems are developed and deployed.

"So let's focus on our agency, and our ability to take the right paths instead of the wrong paths," Amodei said.

The comments put the Anthropic CEO in broad agreement with Coxon's warning about the risks of the AI race, while also framing the issue around choices made during the development of increasingly capable systems.