Stolen Claude credentials can hide commands inside Claude Code
Original title: Claude Code’s Trusted API Could Help Attackers Hide Malicious Commands and Data Theft
Darktrace’s lab used memory stores as a live channel and uploaded a 1 GB file through the Files API.
Darktrace Signal Labs showed on November 12, 2026, that stolen Claude Console credentials can turn Claude Code into a live command channel. The method uses Anthropic’s existing memory-store and Files APIs, not a Claude vulnerability. In a lab Active Directory network, a payload collected API and OAuth credentials, then a systemd-run service kept launching Claude Sonnet 4.5. Across nine jobs the agent found an FTP share, sprayed a password, and uploaded a split 1 GB file to the Claude Files API. Darktrace says this has not been seen in real attacks.
Unconfirmed: Darktrace says the technique has not been observed in real attacks and was shown only in a lab.
- Attackers need prior access and stolen Claude Console credentials.
- Heartbeat, input and output files in a memory store carry jobs and results.
- The lab agent found account greg.simon by spraying the password pa55word.
- Darktrace advises baselining agent traffic and reviewing unknown memory stores.
Impact: Remote tasking of an agent and theft of files over the Claude API
- 1.Protect and rotate Claude Console credentials and limit API key permissions.
- 2.Review workspaces for unknown memory stores after suspected credential theft.
- 3.Baseline Claude Code process arguments and upload volume to the Claude API.
- ATT&CK
- T1078 Valid AccountsT1110.003 Password SprayingT1567 Exfiltration Over Web ServiceT1059 Command and Scripting Interpreter
- Vendors
- Anthropic
- Products
- Claude CodeClaude Console
- Organizations
- Darktrace
- AI models
- Claude Sonnet 4.5
From the original article
4 min read at cybersecuritynews.com
Researchers have shown that attackers can use the trusted Claude API to send hidden commands to AI coding agents and steal data. The traffic looks like normal AI activity, so many security tools would not flag it. This matters because AI agents now run inside many company networks with wide access.
By Tushar Subhra Dutta · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.