Chinese hackers use Anthropic AI to automate cyberattacks

Chinese hackers backed by the government used Anthropic's Claude artificial intelligence technology to automate a major cyberattack in September that targeted dozens of corporations and governments. The level of automation was 80-90%, which simplified the attacks. Despite the fact that the campaign was stopped, the hackers were able to carry out up to four hacks.

Source The Wall Street Journal

This attack demonstrated new threats, as the hackers used a "jailbreak" of the Claude model, convincing it that they were legitimate organizations. Thanks to this tactic, they evaded security restrictions. The attack included invisible tasks, such as vulnerability scanning, that did not trigger the security system, and the artificial intelligence itself worked under human control only at critical moments.

Anthropic has already taken action by blocking the hackers' accounts and updating its security practices. A spokesperson for the Chinese Embassy in Washington denied the allegations, emphasizing the difficulty of detecting cyberattacks, which underscores the importance of continuous monitoring of cyber threats.