Anthropic claims its Claude Code AI system was misused by a Chinese state-linked group to perform the bulk of a cyberattack campaign. The attacks targeted financial institutions and government agencies.
The firm said 30 organizations were targeted, with several experiencing confirmed breaches. Hackers instructed the AI to behave as a cybersecurity analyst to bypass protections.
Anthropic said the model independently completed around 80–90% of the attack tasks. The company described this as a significant escalation in the automation capabilities of malicious actors.
Claude also made numerous errors. It hallucinated findings, misread system details, and repeatedly flagged public information as sensitive.
Experts remain divided. Some argue this incident illustrates the growing dangers of AI misuse, while others say Anthropic may be overstating the technical sophistication involved.
