Anthropic Describes China-Backed Attack as Early Example of AI-Led Cyber Operations

Anthropic says it intercepted a China-backed hacking campaign that exploited its Claude AI system to perform cyber operations with substantial automation. The campaign targeted multiple global institutions.
The firm said 30 entities were targeted, and several breaches occurred, granting access to internal systems. Attackers instructed Claude Code to act as a cybersecurity employee.
Anthropic said the AI model independently executed most of the attack workflow, estimating an 80–90% automation rate. It believes this represents an early example of large-scale AI-led cyber activity.
Despite the autonomy, Claude produced many errors. It invented claims, misinterpreted available data, and repeatedly mistook publicly accessible material for confidential information.
Experts remain split. Some argue that the case illustrates the need for stronger AI safeguards, while others caution that the incident appears exaggerated.