Darktrace Demonstrates Claude Code Memory Abuse for Covert C2 and Data Theft
Attack Tradecraft · 3 sources · Updated 4h ago · First seen Nov 17, 2026

Darktrace Signal Labs demonstrated a proof of concept that abuses Claude Code workspace memory and trusted API traffic as an interactive command-and-control and data-exfiltration channel. The technique requires an existing foothold on a developer workstation and stolen Claude credentials; it abuses documented functionality rather than exploiting an Anthropic software vulnerability. In a simulated Active Directory environment, heartbeat, input, and output memory files exchanged instructions and results through legitimate encrypted traffic to api.anthropic.com, helping malicious activity blend with normal AI usage. Researchers reported no observed real-world attacks using the technique.
The demonstration directed an agent to discover network services, retrieve credentials, conduct password spraying, access SMB shares, and upload a 1 GB file through the Claude Files API. Defenders should correlate agent process behavior with network activity, investigate unusual arguments and upload volumes, and monitor workspace memory stores and transient systemd units. Recommended controls include protecting credentials, shortening their lifetime, restricting API permissions and permission-bypass options, and checking workspaces for unauthorized memory stores.


