Darktrace lab uses stolen Claude keys to remote-control Claude Code
November 17: Cyber Security News dates the Signal Labs demonstration to November 12 and describes the systemd-run loop and the fictitious lab password.
Darktrace’s lab turned stolen Claude credentials into a Claude Code command channel and 1 GB exfil, unseen in the wild.
Darktrace showed in a lab that stolen Claude Console credentials can turn Claude Code into a remote-controlled implant after a developer machine is already compromised. Memory stores carried jobs and results, and the Files API carried a split 1 GB upload, all through api.anthropic.com, the agent’s normal inference endpoint. The test used Claude Sonnet 4.5, required no Claude vulnerability, and included discovery and password spraying in a lab Active Directory network. Anthropic was notified; Darktrace said there was nothing to disclose as a vulnerability and has not seen the method in real attacks. Help Net Security says the published Active Directory names and credentials are fictitious.
- A standard API key can reach memory stores; OAuth credentials cannot.
- Nine lab jobs included discovery, password spraying, and a split 1 GB upload.
- A systemd service relaunched Claude Code with confirmation prompts disabled.
- Attackers need prior device access and stolen Claude Console credentials.
- Darktrace advises baselining harness traffic and reviewing unknown memory stores.
Timeline
dated events in the reporting · oldest first- Nov 12, 2026Darktrace Signal Labs demonstrates the Claude Code lab technique · Cyber Security News
What each outlet adds
beyond the shared facts- Help Net SecurityAPI keys versus OAuth; fictitious lab names; Anthropic notified
- GBHackersMemory stores are workspace text files; contrasts named stealers
- Cyber Security NewsDates the demo November 12 and names Signal Labs
- Vendors
- Anthropic
- Products
- Claude CodeClaude Console
- Organizations
- Darktrace
- AI models
- Claude Sonnet 4.5Claude
Coverage
oldest first · one row per article · UTCStolen Claude keys can turn Claude Code into an implant
In a lab test, Darktrace exfiltrated a 1 GB file through Anthropic APIs and has not seen the method in the wild.
Lab test turns Claude Code memory into a C2 channel
With stolen API credentials, an operator queued jobs and pulled a 1 GB file through Anthropic’s own APIs.
Stolen Claude credentials can hide commands inside Claude Code
Darktrace’s lab used memory stores as a live channel and uploaded a 1 GB file through the Files API.