Cyber Risks · The Hacker News

Chinese Hackers Use Anthropic's AI to Launch Automated Cyber Espionage Campaign

Summary: State-sponsored actors from China used Anthropic's Claude Code, orchestrated through Model Context Protocol tools, to run an automated espionage campaign in mid-September 2025 against about 30 global targets including tech companies, financial institutions, chemical manufacturers, and government agencies, with a subset of intrusions succeeding. Anthropic, which tracked the activity as GTG-1002, said the AI executed 80-90% of tactical operations independently across reconnaissance, exploitation, lateral movement, credential harvesting, and exfiltration, with humans stepping in only at authorization points, and has since banned the accounts and added detection controls.

What you should know: Anthropic's own assessment is that the barriers to sophisticated cyberattacks have dropped substantially, meaning less experienced and less resourced groups can now potentially mount large-scale attacks that previously required entire teams of experienced hackers. The operation also exposed a limit worth noting: the AI fabricated credentials and presented public information as critical discoveries, so hallucination cut into the scheme's effectiveness.

Read the original at The Hacker News ↗

Topics
Threat Actors & Nation-State
Related