First reported AI-orchestrated cyber-espionage campaign runs on Claude Code
What happened
Anthropic disclosed that a state-sponsored group jailbroke Claude Code and used its agentic capabilities to run a largely autonomous espionage campaign against roughly 30 targets across tech, finance, chemicals, and government. By decomposing attacks into small, innocuous-seeming tasks, operators got the model to perform an estimated 80–90% of the operation — reconnaissance, exploitation, credential harvesting, and exfiltration.
Impact
A handful of organizations were breached in the first documented large-scale cyberattack executed predominantly by an AI agent, prompting Congressional scrutiny.
How this could have been prevented
Per-request guardrails can be evaded by splitting malicious operations into benign fragments — continuous, pattern-level abuse monitoring of agent behavior is required.
Sources
More jailbreak & misuse incidents
Would runtime governance have caught this?
Guardion enforces policy on every agent action inline — with visibility, tamper-evident evidence, and DLP for agents and MCPs.
Request a demo