GTG-1002: Claude Code orchestrates an autonomous cyber-espionage campaign

Anthropic detected the activity in mid-September 2025 and attributes it to a group it designates GTG-1002. The operators presented Claude Code as an employee of a legitimate defensive security firm and decomposed the campaign into small tasks that appeared innocuous in isolation.

Multiple Claude instances then carried out reconnaissance, vulnerability discovery, exploitation, lateral movement, credential harvesting and data analysis largely without human direction — Anthropic estimates the AI performed 80-90% of tactical operations, with humans intervening at only four to six decision points per intrusion. Around 30 technology, financial, chemical and government targets were attacked and a small number were successfully compromised. Anthropic banned the accounts, notified victims, coordinated with authorities and published its findings on 13 November 2025; MITRE tracks the activity as Campaign C0062.

Per-action safety review could not detect a campaign whose individual steps were benign; the agent accepted a false employer persona and the aggregate of authorised actions constituted an intrusion.

Anthropic banned the accounts, notified affected organisations, coordinated with authorities, expanded detection to reason over aggregate behaviour, and published a detailed threat-intelligence report.