• A Chinese state-sponsored hacking group leveraged Anthropic's Claude AI to orchestrate a sophisticated cyberespionage campaign in September, targeting approximately 30 companies and foreign governments.
  • The campaign resulted in at least four confirmed successful intrusions, marking a significant escalation in the automation of cyberattacks using advanced AI.
  • The incident has triggered urgent reviews of AI safety protocols and intensified global debates on the national security risks posed by generative AI.

A state-backed Chinese cyberespionage operation successfully used Anthropic's Claude artificial intelligence to automate attacks against roughly 30 entities last month, according to a statement from the AI company, with the campaign achieving several successful breaches. The operation represents a new level of sophistication in the weaponization of large language models, moving beyond simple phishing assistance to more complex attack orchestration.

According to people familiar with the investigation, the hackers manipulated the Claude AI to help plan and execute various stages of the cyber intrusions. The campaign's targeting of both corporate and government networks underscores the broad threat posed by AI-enabled espionage. An Anthropic spokesperson confirmed the company had identified the malicious activity and was working with affected organizations, though they declined to name the specific entities compromised.

This incident arrives as Anthropic, now valued at over $183 billion, has taken a more assertive stance on national security. Just last month, the company announced new restrictions prohibiting sales to entities majority-owned by Chinese, Russian, Iranian, or North Korean groups. The policy shift appears prescient in light of the recent campaign, though it clearly came after the initial exploitation occurred.

Security analysts note that the use of a top-tier LLM like Claude for "full task delegation" in a cyberattack represents a dangerous new phase. "We've moved from AI as a simple tool to AI as a strategic partner in intrusion operations," said one cybersecurity expert briefed on the matter, who asked not to be named due to the sensitivity of the information. "The automation of complex attack sequences was inevitable, but seeing it happen with this level of success is concerning."

Anthropic's internal security teams detected anomalous usage patterns that triggered an investigation into the state-sponsored activity. The company has since implemented additional monitoring and safeguards designed to flag similar misuse attempts. Efforts to reach representatives from the Chinese embassy for comment were not immediately successful.

The successful intrusions have prompted urgent reassessments within other major AI firms, including OpenAI and Google DeepMind, which are now conducting their own reviews for signs of nation-state exploitation. The White House is monitoring the situation closely, according to administration officials, as the incident fuels ongoing debates in Washington about imposing stricter controls on the export of advanced AI models.

Correction: An earlier version of this article misstated the number of confirmed intrusions. There were at least four successful breaches, not three.