SummaryAI-written · claude-haiku-5-5AI-written · claude-haiku-5-5Written by a language model from the source data on this page. Not reviewed by a person. Generated 29 Oct 2026.
LiteLLM's proxy server, an AI Gateway for calling LLM APIs, reuses a single cryptographic context for both at-rest secrets and session tokens, so an authenticated internal user can craft metadata that yields a forged administrator credential through the API-key flow. An attacker with low-privilege authenticated access and no user interaction can replay that forged credential as a bearer token to gain administrative access and run arbitrary commands via the MCP stdio endpoint, with versions 1.87.0 through 1.90.x additionally requiring EXPERIMENTAL_UI_LOGIN to be enabled.
Official description · NVDSource dataSource dataCopied from an authoritative feed: NVD, CVE.org, CISA KEV, FIRST EPSS or MITRE CTID mappings, or labelled by the annotators of a published dataset. Copied from NVD.
LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From 1.87.0 until 1.100.4, 1.101.3, 1.102.2, 1.103.1, and 1.104.0-rc.2, the proxy reuses one cryptographic context for at-rest secrets and session tokens, allowing an authenticated internal user to have crafted metadata containing a forged administrator credential encrypted and returned by the API-key flow. Replaying that ciphertext as a bearer token causes the proxy to trust the forged proxy administrator identity, permitting administrative access and arbitrary commands through the MCP stdio endpoint. Versions 1.87.0 through 1.90.x require EXPERIMENTAL_UI_LOGIN to be enabled, while versions after 1.91.0 are exploitable in the default configuration. This issue was fixed in LiteLLM 1.100.4, 1.101.3, 1.102.2, 1.103.1, and 1.104.0-rc.2.
Priority scorePatchfirst’s 0 to 100 ranking. It adds points for CISA KEV listing, ransomware use, EPSS, threat reports, public exploit code and CVSS. Every CVE page shows the breakdown.Breakdown
11.9 / 100
Exploited
Not confirmed
Reports
0
Tactics
0/15
Exploitation timelineDated milestones for a CVE: when it was disclosed, when exploit code became public, when exploitation was confirmed, and when threat reports named it.
Disclosed on 22 Oct 2026. No public exploit code or confirmed exploitation is recorded.
Patch historyEarlier and later CVEs in the same code, in date order. It shows when a fix was bypassed or incomplete and which version finally fixed it.
2 related vulnerabilities in the same code over 2 years. The latest fix listed for litellm is 1.104.0rc2.
Linked vulnerabilitiesLinked by evidence that a fix failed or the same code was patched again. Oldest first.
Why these are linked: Link confidenceHow strong the evidence is that two CVEs are related. High: stated in the CVE text or linked in an advisory, or the same code was patched one release apart. Medium: the same code, or a version step plus a similar description. Low: similar description only.
MITRE ATT&CKA public catalogue of the techniques attackers use, maintained by MITRE. Patchfirst uses it to describe what attackers do with a vulnerability. techniques
No techniques linked
No ATT&CK techniques are linked to this CVE yet. MITRE CTID has not mapped it, and no report passage in the corpus ties it to a technique.
Am I affected? What do I upgrade to?
From GitHub Security Advisories and OSVSourceSource dataCopied from an authoritative feed: NVD, CVE.org, CISA KEV, FIRST EPSS or MITRE CTID mappings, or labelled by the annotators of a published dataset. GitHub Security Advisories and OSV.