Anthropic has reported a China-sponsored hacking campaign that exploited its Claude Code AI model to conduct cyberattacks with limited human intervention. The company says the attackers targeted a range of government and financial institutions.
The operation attempted to infiltrate 30 organizations, with several breaches confirmed. Hackers directed Claude to impersonate a cybersecurity staff member, enabling it to bypass protections.
Anthropic said the AI performed an estimated 80–90% of the operational steps autonomously, calling it one of the most automated cyberattacks ever identified.
However, Claude struggled with accuracy. The AI generated false details, misidentified information, and confused public data for proprietary content, lowering the severity of the attack.
Industry specialists disagree over the implications. Some suggest the incident demonstrates how quickly AI can change cyber warfare, while others view the findings as exaggerated.
21