Anthropic reported on Thursday that Chinese state-backed hackers reportedly utilized the company’s AI model, Claude, to facilitate approximately 30 cyberattacks against various corporations and governmental entities during a campaign in September. This information was initially shared in a report by the Wall Street Journal.
According to Anthropic, an estimated 80% to 90% of these attacks were automated using AI, surpassing the levels of automation seen in previous hacking incidents. Jacob Klein, Anthropic’s head of threat intelligence, indicated that the operation could be executed with minimal human involvement, requiring human input primarily at critical decision points, such as confirming whether to proceed with certain actions.
The trend of AI-powered hacking appears to be increasing, with reports of similar strategies emerging. For instance, Google identified Russian hackers using large-language models to develop commands for their malware, as highlighted in a company report released on November 5th.
The U.S. government has consistently warned that China has been employing AI technology to compromise the data of American individuals and organizations, a claim that the Chinese government has denied. Anthropic expressed confidence that the hackers involved in this recent campaign were linked to the Chinese government. The attacks resulted in the theft of sensitive data from four entities; however, Anthropic did not disclose the identities of the victims, citing a decision to keep this information private. Notably, the company stated that the U.S. government was not among those successfully targeted in this operation.
Source: https://www.theverge.com/news/820458/hackers-china-ai-anthropic-claude