CISA says attackers are using a LiteLLM sandbox escape. The AI proxy runs as root.
By Lior AharonovFounder, 7IT Solutions ·
What happened
On 18 November 2026 CISA added CVE-2026-40217 in BerriAI LiteLLM to its exploited-vulnerabilities list, with a due date of 2 December. The flaw lets code escape a sandbox on the POST /guardrails/test_custom_code endpoint. The vendor advisory says the default Docker image runs the proxy as root.
Why it matters
A LiteLLM proxy often sits in front of every model key your team owns. Code running inside it can reach whatever the proxy can reach. The advisory says exploitation needs a proxy-admin credential in default setups, so a leaked admin key matters more now.
What to do
Upgrade LiteLLM to 1.83.11 or later today. If you cannot, block POST /guardrails/test_custom_code at your reverse proxy, and check who holds proxy-admin keys.

My take
Lior Aharonov
Founder, 7IT Solutions
An AI gateway is a root-run web service holding your keys. Patch it like one, not like a dev toy.
Sources
| CISA adds four known exploited vulnerabilities to catalog | cisa.gov |
| LiteLLM security advisory GHSA-wxxx-gvqv-xp7p | github.com |
Researched with AI tools; every fact is checked against the linked sources.