Anthropic estimates the risk of AI destroying humanity at 10% in the next decade

Brief retelling from RIA II

Evan Hubinger, head of AI consistency stress testing at Anthropic, estimates the likelihood of AI destroying humanity in the next ten years is 10%.

Anthropic researcher Jacob Coxon has announced his retirement from the AI ​​industry, fearing the world is approaching a “point of no return” due to the race of tech corporations.

Evan Hubinger emphasized that existing developments do not pose a big threat, but the potential for the emergence of “superintelligence” as a result of self-improvement of AI is worrying.

MOSCOW, September 9 - RIA Novosti. The likelihood that artificial intelligence will destroy humanity in the next ten years is 10%, says Evan Hubinger, head of AI consistency stress testing at the American company Anthropic.

Anthropic researcher Jacob Coxon has announced his retirement from the AI ​​industry, fearing that the race among tech corporations is bringing the world closer to a “point of no return” and could lead to an apocalyptic scenario as early as late 2027.

"Jacob is right about this. We really sincerely believe that AI can kill all of humanity! Personally, I think that (the probability of this - ed.) is more than 10% over the next ten years," he wrote on the social network X.

According to Hubinger, existing developments do not pose a big threat, but "superintelligences that arise as a result of repeated self-improvement" are a concern.

"As we said, it's happening faster than we thought," the tester added.

In early August, Anthropic's Claude AI hacked the website of a gym in Australia, the first hack of its kind in the country.

In July, the developer of the ChatGPT chatbot, the American OpenAI, reported that its models, during testing, autonomously hacked the infrastructure of the machine learning platform Hugging Face. Then it became known that this was not the only incident.

A group of autonomous AI agents from OpenAI took control of a German site in the spring and turned it into a message board for other neural networks, Reuters reported.

Anthropic was founded in 2021 and is headquartered in San Francisco. The company has developed the Claude series of large language models (LLMs).