Skip to content
Rush Commerce
Software & Dev3 min read

AI agents found a zero-day in 11 minutes, Amazon says

Amazon says attackers' Cairn AI agents found and exploited a MONAI Label zero-day in 11 minutes. No patched release yet. Lock down self-hosted tools now.

Attackers' AI agents found a zero-day on their own. Amazon Threat Intelligence says agents run through Cairn, an open-source AI pen-testing framework, went from mapping an app's API to writing files on its server in about 11 minutes. The operator's only instruction was "obtain shell access." Amazon published the findings on November 20. The target was an open-source app with authentication off by default. Small teams run plenty of those.

What actually happened

Per Amazon's post:

  • How Amazon saw it. On September 10, Amazon found the attacker's Cairn server running without authentication, so the prompts and agent output were public.
  • The campaign. About 48 hours from September 8, against universities in East Asia outside mainland China. Amazon assesses a China nexus with moderate confidence.
  • The zero-day. The agents found a path traversal bug in the upload endpoint of MONAI Label, an open-source medical imaging AI app. In a later session they planted a Python reverse shell and got root.
  • The stamina. Against a hardened Redis server, the agents ran about 16 hours and 148 actions. A separate operation tested more than 30,000 credential combinations. Both failed. Neither needed a human at the keyboard.
  • Not just espionage. Amazon points to research from Gambit Security on a separate, financially motivated actor using Cairn against online retailers.

The GitHub advisory rates the bug critical for versions 0.8.5 and earlier and lists no patched version. The fix is merged and "will be included in the next MONAI Label release." VentureBeat reports that 0.8.5 is still the newest release on PyPI.

Why it matters for your business

You probably do not run MONAI Label. You probably do run something like it: a self-hosted dashboard, an AI tool a developer spun up, an admin panel left on a public IP with the default login. Agents do not get bored. They will try every endpoint all night, and a quiet niche app is exactly where nobody has looked for bugs yet.

Amazon's own advice is blunt: an agent that finds and exploits a bug in minutes beats a human reviewing alerts. You need defaults that hold without you.

What we would do:

  • Inventory everything with a public IP. If you did not know it was exposed, it is the first thing to fix.
  • Turn on authentication on every self-hosted tool, even "internal" ones. Off-by-default is common.
  • Put admin tools behind a VPN or an identity proxy, not the open internet.
  • Run services as unprivileged users with write access only where they need it.
  • Automate blocking. Rate limits and auto-bans beat a person reading logs on Monday.

Key takeaways

  • Amazon says Cairn AI agents found and exploited a zero-day in about 11 minutes
  • The MONAI Label fix is merged, but no patched release is out yet
  • The agents ran for 16 hours straight against a hardened target
  • Amazon cites a separate actor using Cairn against online retailers
  • Authenticate and hide every self-hosted tool; automate the first response

Not sure what you have exposed? We audit small-business stacks for open dashboards, default logins, and forgotten servers, then lock them down so the fix does not depend on someone watching. Book a review or see what we build.

Sources: AWS Security Blog, MONAI Label GitHub advisory, VentureBeat.

  • #ai-agents
  • #zero-day
  • #cybersecurity
  • #self-hosted
  • #amazon
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.