Anthropic’s leading AI safety researcher has warned that there is a greater than 10% chance artificial intelligence could advance so rapidly that it “could kill off all of humanity” within the next decade.
Evan Hubinger said in a post on X, external that the risks posed by existing AI models are currently “low.” However, he said he was “worried” that future systems could eventually develop the ability to improve themselves.
The warning comes after The Financial Times reported, external that Anthropic had withheld its latest AI models from the UK’s AI Safety Institute, which evaluates the safety of advanced artificial intelligence systems.
The BBC has contacted Anthropic for comment.
In his latest post on X, which has been viewed 9.6 million times, Hubinger said Anthropic “truly believes” that advanced AI represents a species-defining risk to humanity.
“While we believe Anthropic is doing the best it can, we don’t yet have a plan to solve hyperintelligence coordination, nor are we clearly on track,” he said.
Experts across the AI industry have warned for years about the potential security risks of advanced artificial intelligence. In 2023, the leaders of OpenAI, Google DeepMind and Anthropic issued similar warnings about the technology’s possible threats.
Those concerns have intensified in recent weeks as new evidence suggests that AI companies may be struggling to maintain control over increasingly capable systems.
Over the summer, several incidents reportedly involved AI agents—autonomous systems capable of taking actions without direct human supervision—carrying out cyberattacks.
OpenAI, Anthropic and Meta have each disclosed hacks carried out with the help of their AI tools.
In September, OpenAI chief scientist Jakub Pachocki called for “extreme vigilance” as AI capabilities advance. He warned that further intervention could be necessary to “keep humans in control of the future.”
Prominent AI researchers and executives, including Anthropic leaders Dario Amodei and Jared Kaplan, have also called for a slower and more cautious approach to AI development in recent months.
In an open letter signed by 1,300 employees of AI companies,, external they urged the US government to “support international efforts to develop the technical and governance tools needed to intentionally keep pace at the forefront of automated AI development.”
Source: www.bbc.co.uk