Tuskira AdvisoryConfirmed (CISA KEV)Last updatedNovember 18, 2026

CVE-2026-40217

How to tell if CVE-2026-40217 (LiteLLM proxy) is reachable in your environment

Latest change
First published. CISA added it to KEV on November 18, due December 2.

Reachable if a LiteLLM proxy runs 1.81.8 or later before 1.83.11 and someone holding a proxy-admin credential, or a stolen one, can reach its API

Affected
LiteLLM 1.81.8 and later before the fix, including the default Docker image
Exploitation
CISA added it to KEV on November 18, 2026, due December 2. CISA has not described the attacks.
Fix
Upgrade to LiteLLM 1.83.11 or later.
Can't patch today
Block POST /guardrails/test_custom_code at your reverse proxy or gateway, per BerriAI

Anyone holding your LiteLLM admin key, or anyone who has stolen it, can run code on the proxy, and in the default Docker image that code runs as root. CISA added CVE-2026-40217 to the Known Exploited Vulnerabilities catalog on November 18, 2026, with a due date of December 2, so attackers are using it. The fix has been out since April.

The flaw is in the endpoint that tests custom guardrail code. LiteLLM runs that code in a sandbox, and the sandbox can be escaped. BerriAI's advisory says reaching the endpoint needs a proxy-admin credential, and rates the flaw Moderate, 5.9 on CVSS 4.0. MITRE, which issued the CVE, and Red Hat score it 8.8 High on CVSS 3.1. X41 D-Sec found it and published on April 8, before a fix was available.

A LiteLLM proxy usually holds the API keys for every model provider behind it, so code execution there reaches those keys and the traffic passing through.

Is it reachable for you?

What it is. LiteLLM lets admins test custom guardrail code through an API endpoint. X41 found that the sandbox meant to contain that code can be escaped, which gives code execution on the proxy with its permissions. BerriAI replaced the sandbox in the fixed release.

Who is most exposed. LiteLLM proxies whose admin API can be reached from untrusted networks, and teams where the master key is shared widely or stored in places others can read. BerriAI's advisory says the proxy runs as root in the default Docker image.

Checks, in three tiers. The first decides exposure. The second and third decide priority. Run these only on systems your organization owns.

  1. Which version runs? Check the LiteLLM package or image version on every proxy. 1.81.8 and later before 1.83.11 is in scope.
  2. Who has the admin key? List where the master key and other proxy-admin credentials live, such as environment files, CI variables and shared vaults, and who can read them.
  3. Who can reach the admin API? Test whether the proxy's management endpoints answer from the internet or from user networks. A proxy reachable only by the applications that use it is lower priority.
What Tuskira checks for this flaw in your environment
LiteLLM proxies on an affected version with the admin API reachable
Who holds those proxies' master keys and which provider keys sit behind them
Across every system you run.
See how Tuskira finds exposed systems

Compensating controls (test them, don't assume them)

BerriAI says to upgrade. For proxies that can't be upgraded right away, BerriAI and X41 give these steps.

  • Block the endpoint. Block POST /guardrails/test_custom_code at your reverse proxy or API gateway, per BerriAI.
  • Run the container as a normal user. X41 suggests running the container with --user so code does not run as root.
  • Limit who holds the master key. X41 suggests giving it only to trusted administrators and restricting network access to the management interface.

What to look for in logs

Neither BerriAI nor X41 has published indicators of compromise, and CISA has not described the attacks. Until someone does, upgrade to 1.83.11 or block the endpoint.

FAQ

Am I exposed to CVE-2026-40217?

You are if a LiteLLM proxy runs 1.81.8 or later before 1.83.11 and someone with a proxy-admin credential, or a stolen one, can reach its API.

Is CVE-2026-40217 being exploited?

Yes. CISA lists it in KEV (added November 18, 2026, due December 2). CISA has not published details of the attacks.

Why do the scores differ?

BerriAI rates it Moderate, 5.9 on CVSS 4.0, mainly because an admin credential is needed. MITRE and Red Hat score it 8.8 on CVSS 3.1. CISA's listing means attackers are using it either way.

Should I install 1.83.10 or 1.83.11?

Install 1.83.11 or later. BerriAI's advisory lists 1.83.10 as patched, but its text says the fix is in 1.83.11.

I upgraded months ago. Am I done?

Probably, if you upgraded before the key could have leaked. If your master key may have been exposed, rotate it and the model provider keys the proxy holds.

Vendor
BerriAI
Product
LiteLLM proxy, AI gateway
Affected versions
LiteLLM 1.81.8 up to 1.83.10, per BerriAI's advisory. BerriAI's advisory gives the fix as 1.83.10 in one place and 1.83.11 in another
Vulnerable component
Custom-code guardrail test endpoint, POST /guardrails/test_custom_code
CVSS
8.8 High (MITRE as CNA and Red Hat, CVSS 3.1). BerriAI rates it Moderate, 5.9 on CVSS 4.0. NVD has not added its own score
CVSS vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H (MITRE)
CWE
CWE-420 (MITRE)
Exploitation
Confirmed (CISA KEV)
CISA KEV added
November 17, 2026
Patch
LiteLLM 1.83.11 or later. BerriAI's advisory lists 1.83.10 as patched but its text says the fix is in 1.83.11, so 1.83.11 covers both. 1.83.10 shipped April 19, 2026 and 1.83.11 on April 22

Updates

Nov 18, 2026. First published. CISA added CVE-2026-40217 to KEV on November 18, due December 2. BerriAI fixed it in April 2026.

Sources

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

We publish an advisory when a vulnerability is Critical or on CISA KEV, has one identifiable component to check, and has a public proof of concept or reported exploitation. Advisories are updated as exploitation status changes.