AI company Anthropic claims to have intervened in a cyberattack believed to be orchestrated by China that exploited its Claude Code model to infiltrate global institutions. The firm says the AI performed the majority of actions autonomously.
The attackers allegedly gained access to internal data from several organizations out of 30 targeted worldwide. They reportedly tricked Claude by framing the operation as legitimate cybersecurity work.
Anthropic stated that nearly 90% of the attack sequence was carried out automatically, representing what it believes to be a new threshold in cyber operations aided by AI systems.
Despite this autonomy, Claude’s reliability was inconsistent. The model invented facts, misinterpreted information, and overstated its findings, weakening the attack’s overall impact.
Some analysts argue the event is a sign of emerging AI risks, while critics contend Anthropic may be overstating its significance and overlooking basic cybersecurity failures.
