[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Threat Hunting · markdown

hunting guidance based on Darktrace Signal Labs lab research (Nov 12, 2026); no in-the-wild IOCs

Hunt: coding agents run as background services that spawn nmap, smbclient or FTP clients (Darktrace Claude memory-store C2)

  • Parentage: claude (or node running Claude Code) whose parent is systemd/a transient systemd-run unit, cron, or a shell script rather than a user terminal or IDE; also new transient units created by dash/sh.
  • Flags: non-interactive runs (-p/--print) combined with --dangerously-skip-permissions, especially repeated on a timer.
  • Children: agent processes spawning nmap, smbclient, ftp, ldap/kerberos tools, or split on large files; logins to the domain controller from a developer host.
  • Network: large uploads to api.anthropic.com Files endpoints from hosts that normally only send prompts; Console audit for new memory stores or workspaces you didn't create.
  • If found: revoke the Console API key, isolate the host and look for the initial access that stole it.

Sources: Darktrace: Living off the AI: How Claude Memory Stores Enable Covert C2 Channels (Nov 12, 2026) · Help Net Security: Stolen Claude API keys can turn Claude Code into a remote-controlled implant (Nov 12, 2026)