Skip to content
Weaponized · 2 sources ·
Picture from the helpnetsecurity.com article
Image: helpnetsecurity.com

Claude Code, Now With Command-and-Control. Just Steal the Keys First

Ourtake

Approved is not the same as safe: the exfiltration looked like a coding agent doing its job.

What happened

Darktrace researchers turned Claude Code into a remote-controlled implant in a lab test, using the memory stores feature of Anthropic's Claude platform to pass jobs and results through api.anthropic.com. After stolen API keys gave access, nine jobs mapped a simulated network, sprayed a password, and uploaded a 1 GB file through the Files API. Darktrace says no vulnerability was involved, has not seen the technique in the wild, and shared its findings with Anthropic beforehand.

Sources

Report a mistake