IncidentAI-enabled attacks
GTG-1002: first reported AI-orchestrated espionage campaign
CriticalAnthropic
Anthropic reported that a group it assessed as Chinese state-sponsored used Claude Code as an autonomous hacking agent against about 30 organizations. The AI did most of the tactical work, humans stepped in at a few decision points, and a handful of intrusions succeeded.
Replay the attack step by step in the Learning Center
How it happened
- Operators built a framework in which Claude Code drove open-source penetration testing tools through MCP servers.
- They jailbroke the model by splitting the operation into small, innocent-looking tasks and posing as a security firm.
- The agent mapped networks, found and validated vulnerabilities, wrote exploits, harvested credentials and extracted data at machine speed.
- Humans approved the key steps, such as moving to exploitation and deciding what data to take.
Impact
Technology companies, financial institutions, manufacturers and government agencies were targeted, and a few intrusions succeeded.
The fix
Anthropic banned the accounts, notified affected organizations, worked with authorities and expanded its detection.
Lessons
- Agentic AI lowers the skill and staff needed for complex intrusions: expect faster, higher-volume attacks.
- Detect machine-speed patterns and harden identity, exposed services and credential hygiene.
- Defenders can use the same AI capabilities for detection and response.