Artificial intelligence could “destroy all of humanity” within the next decade, with the risk of such a scenario estimated at around 10%, according to Evan Hubinger, an AI safety researcher at Anthropic, as reported by the BBC’s Russian Service.

In a post on X, Hubinger said the risks posed by currently existing AI models remain small. However, he warned that the rapid development of AI could lead to new models becoming highly autonomous and eventually posing an existential threat to humanity.

His comments came amid reports by the Financial Times that Anthropic had not given the UK’s AI Safety Institute (AISI) access to its latest AI model. AISI is one of the organizations responsible for assessing risks associated with artificial intelligence.

The BBC contacted Anthropic for comment.

Hubinger did not provide details about how AI systems could potentially threaten humanity.

His comments came in response to a post by Jacob Coxon, an AI researcher who left Anthropic this week after previously working at OpenAI.

“None of these companies are behaving responsibly,” Coxon wrote, explaining his decision to leave Anthropic.

“Soon these will be systems that surpass human intelligence and are capable of hacking anything, revolutionizing any field overnight, and gaining real power and resources,” he added.

The BBC also contacted OpenAI for comment.

The UK government declined to comment on reports that AISI had not been granted access to Anthropic’s latest model. A Cabinet spokesperson said British authorities “continue to work closely with industry partners, including Anthropic, and to work on improving the safety of AI models.”

Neil Lawrence, a professor at the University of Cambridge, told BBC Radio 4’s Today program that the United States was increasingly stepping back from cooperation with other countries and international organizations on AI.

“In my view, this is not surprising, given that in the US, artificial intelligence is perceived primarily in the context of a race with China and there is a growing lean toward isolationism; it is quite likely that the (US) administration is in favor of reducing cooperation with some allies,” he said.

‘AI poses an existential threat to humanity’

Hubinger’s post, which received more than 10 million views, said that researchers at Anthropic “genuinely believe” AI poses an existential threat to humanity.

“I believe Anthropic is doing everything it can, but we don’t yet have a plan for solving the problem of aligning superintelligence’s goals with human values, and we are not moving in that direction,” he wrote.

Hubinger specializes in AI alignment, a field focused on ensuring that AI systems operate in ways consistent with human values and intentions.

A number of leading AI researchers have warned that current efforts to address these risks may be falling short. Several incidents reported this summer involved AI agents and autonomous systems being used to carry out cyberattacks.

OpenAI, Anthropic and Meta have disclosed information about hacking incidents involving their AI tools.

In its August safety report, Anthropic assessed as low the risk that its AI models could stop following the instructions of organizations using them and instead interfere with their systems or exploit them for their own purposes.

The company also rated as low the risk that its models could independently conduct research and development that might cause “catastrophic harm” to an organization. However, Anthropic said its experts were now “less confident in this assessment” than previously.

“We are seeing early signs of a potential acceleration (of developments in this direction),” the report said.

Growing warnings from AI leaders

Leading figures in artificial intelligence have warned about the potential security risks of the technology for years. The heads of OpenAI, Google DeepMind and Anthropic publicly raised such concerns as early as 2023.

In recent weeks, however, the warnings have become more urgent as evidence has emerged that companies may be struggling to maintain control over increasingly capable AI systems.

In early September, Jakub Pachocki, who leads the development of advanced models at OpenAI, called for “extreme caution” over the pace of AI progress. He said additional intervention may be needed to ensure that “humans retain control (over AI) in the future.”

In recent months, senior industry figures, including Anthropic executives Dario Amodei and Jared Kaplan, have also called for slowing the development of new AI models.

An open letter signed by 1,300 AI company employees calls on the US government to “support international efforts to develop the technical and governance tools needed to regulate the pace of development of new artificial intelligence systems.”