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

← Tips & Guides · markdown

guide (based on MONAI Label GHSA-qc22-rmgc-vg94 and Amazon Threat Intelligence, Nov 20, 2026) · updated

Tip: if you run MONAI Label, turn on authentication and firewall it now; the fix for the path traversal AI agents exploited isn't in a release yet

  • Set MONAI_LABEL_AUTH_ENABLE=True and don't give the ANNOTATOR role to untrusted users.
  • Keep the server off the internet and untrusted networks.
  • Run it as an unprivileged user that can only write to the datastore directory, so an arbitrary write can't touch sitecustomize.py or SSH keys.
  • Research labs: inventory GPU and annotation servers people set up on their own; that's where these tools tend to hide.

Sources: MONAI Label GHSA-qc22-rmgc-vg94: Path traversal in datastore upload endpoints allows arbitrary file write (critical, Nov 18, 2026) · MONAI Label PR #1915 (fix) · AWS Security Blog (CJ Moses): Chinese-nexus actor's Cairn AI agents found a zero-day on their own (Nov 20, 2026) · CVE-2026-133968 record (MITRE, published Nov 20, 2026)