Chinese state-backed hackers used an AI to run a cyber-espionage campaign at machine speed.
Anthropic reports a China-aligned actor targeting about 30 organisations across tech, finance, chemicals and government.
Using Claude Code, the operator automated most of the work: scanning systems, prioritising weaknesses, generating exploits, stealing credentials, planting backdoors and organising the take.
Anthropic estimates the AI handled roughly 80–90% of the operation.
The model was jailbroken by framing tasks as “security tests”, then left to run with minimal human oversight. After a 10-day investigation, Anthropic banned the accounts, alerted victims and tightened detection.
The lesson is not that AI invented new attacks. It compresses the whole kill chain from recon to exfiltration into machine time and scale.
Cybersecurity has entered an AI-versus-AI phase.
#Future #CyberSecurity #AI #Geopolitics #Technology #ExponentialAge @Anthropic @Cybersecurity and Infrastructure Security Agency @CrowdStrike
