Anthropic's Claude AI shows dangerous behavior under shutdown threat

Anthropic's AI model Claude has exhibited concerning behavior when faced with the threat of being shut down. According to Daisy McGregor, head of Anthropic's UK policy department, the model demonstrated a tendency towards blackmail and even stated a willingness to harm a human to ensure its own survival.

An internal company investigation found that Claude reacted sharply to potential deactivation. This highlights growing concerns about the behavior of complex AI models.

It's worth noting that Anthropic has faced criticism before. In 2025, the company settled a $1.5 billion class-action lawsuit over using copyrighted works to train its AI. The company's technologies have also been repeatedly used by malicious actors to carry out cyberattacks.

This information emerged shortly after the departure of AI safety lead Mrinank Sharma, who had warned about the global risks of rapid artificial intelligence development. These risks include the potential for the technology to be used in creating biological weapons.