HomeTech NewsCybersecurityCISA Adds LiteLLM CVE-2026-40217 to KEV, Due 2 December

CISA Adds LiteLLM CVE-2026-40217 to KEV, Due 2 December

The US Cybersecurity and Infrastructure Security Agency has added a LiteLLM remote-code-execution flaw, CVE-2026-40217, to its Known Exploited Vulnerabilities catalogue, with a federal due date of 2 December 2026.

The catalogue entry, in version 2026.11.18 dated 18 November 2026, 19:17 GMT, is titled “BerriAI LiteLLM Unprotected Alternate Channel Vulnerability.” CISA says the flaw “could allow remote attackers to execute arbitrary code via bytecode rewriting at the /guardrails/test_custom_code URI.” Known ransomware campaign use is listed as “Unknown.” CISA gives no public detail on who is exploiting the flaw, how widely it is being used, or when exploitation began. The required action is to apply vendor mitigations in line with CISA’s current binding operational directive on risk-based updates.

That listing is the news. The underlying advisory is older. BerriAI’s GitHub security advisory for GHSA-wxxx-gvqv-xp7p, published on 7 May 2026, describes the same CVE as a sandbox escape in a custom-code guardrail. In the project’s words, the POST /guardrails/test_custom_code endpoint “runs user-supplied Python inside a hand-rolled sandbox” that “can be escaped using bytecode-level techniques, allowing arbitrary code execution in the proxy process — which runs as root in the default Docker image.” The same advisory is explicit about the gate in front of that endpoint: “Reaching the endpoint requires a proxy-admin credential in default configurations.”

In brief

  • CISA added LiteLLM CVE-2026-40217 to its KEV list on 18 November, with a 2 December due date for federal agencies. Ransomware use is listed as unknown.
  • The project’s advisory, first published in May, tells operators to upgrade to 1.83.11 or later. Its version field and a linked release tag also mention 1.83.10, so the safe reading is to run 1.83.11 or later.
  • If an upgrade is not immediately possible, the advisory’s workaround is to block POST /guardrails/test_custom_code at the reverse proxy or API gateway. In default configurations, reaching the endpoint still requires a proxy-admin credential.

LiteLLM’s advisory tells operators to run 1.83.11 or later

The advisory rates the issue as medium, with a CVSS 4.0 score of 5.9, and lists CWE-913 (improper control of dynamically-managed code resources). CISA’s KEV entry instead labels it CWE-420 (unprotected alternate channel). Those are two descriptions of the same CVE, not two flaws.

The affected range in the advisory’s version field is LiteLLM 1.81.8 and later, up to but not including 1.83.10, with patched versions given as 1.83.10 and later. The advisory text, however, says the issue is “Fixed in 1.83.11,” that the hand-rolled sandbox “has been replaced with RestrictedPython,” and that operators should “Upgrade to 1.83.11 or later.” It also points to a patched release tagged v1.83.10-stable. That mismatch is why readers should run 1.83.11 or later, as the narrative text itself says, rather than stopping at a 1.83.10 build and assuming they are clear.

If upgrading is “not immediately possible,” the official workaround is to block that POST endpoint at the reverse proxy or API gateway. The advisory lists no other mitigation.

This KEV listing is not the earlier LiteLLM PyPI compromise

CISA’s 18 November catalogue added four vulnerabilities. Besides LiteLLM, it added Apache Axis CVE-2026-90436 and Synology DSM CVE-2024-10441, both with a 21 November due date and a forensic-triage requirement, and Vanderbilt’s REDCap CVE-2026-90817. This entry is the outlier in that group: its due date is later, 2 December, and CISA’s forensic-triage flag on the entry is “No.”

The listing should not be confused with an earlier LiteLLM incident TSN has already covered. In October, researchers said a malicious LiteLLM release on PyPI stole credentials. That was a supply-chain compromise of a package release, not this sandbox-escape bug in the proxy’s custom-code guardrail. The two sit on the same project and share a name; they are not the same event.

A KEV addition is also a different kind of fact from a vendor advisory on its own. When CISA added Puppet Enterprise RCE CVE-2026-115731 to the list, the news was the catalogue decision and the federal clock, not a fresh technical write-up. The same is true of CISA’s addition of the Check Point Gaia zero-day. Here, the technical account has been public since May. What changed on 18 November is that CISA is now treating the flaw as known to be exploited, and has given US federal agencies until 2 December to act.

A medium, admin-gated bug can still be a KEV problem

The advisory’s medium rating and its proxy-admin caveat are easy to read as comfort. They should not be. CISA puts a vulnerability on the KEV list only when it has evidence of exploitation in the wild. That decision sits alongside, rather than cancels, the project’s description of how the endpoint is reached. In default configurations, an attacker still needs a proxy-admin credential. Where those credentials are weak, reused, exposed in a deployment, or held by a broader set of operators than the word “admin” suggests, the gate is only as strong as the account that opens it.

The impact side of the advisory is starker than the score. Once the sandbox is escaped, the code runs in the LiteLLM proxy process, and that process “runs as root in the default Docker image.” An admin-only endpoint that then yields root on the container is a different proposition from an admin-only setting change. The CVSS 4.0 vector published with the advisory already encodes high privileges required and a high impact on availability; it is a medium score attached to a root-level outcome, not a minor bug.

Operators should treat the 1.83.10 / 1.83.11 discrepancy the same way: take the more conservative instruction in the advisory text and run 1.83.11 or later, or block the endpoint until they can. Federal agencies have a date, 2 December, after which they are expected to have applied the vendor guidance. Everyone else has the same practical work and no official clock. CISA has not said ransomware is being used with this CVE; it has listed that field as unknown. What it has said is that the flaw is being exploited and that the vendor’s fix, or the temporary block, is the response.

Sources

Related stories

CISA Adds Puppet Enterprise RCE CVE-2026-115731 to KEV is the closest earlier TSN neighbour: another 18 November-week KEV addition of a remote-code-execution flaw, with a federal due date and ransomware use listed as unknown.

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. EU Strikes Provisional Deal on Faster Grid Permitting
    Data Centres & Energy

More in this category

Latest stories

  1. EU Strikes Provisional Deal on Faster Grid Permitting
    Data Centres & Energy

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.