// Security Advisory Triage
LiteLLM privilege escalation via salt key reuse patched in five versions
GitHub advisories — pip · published 2026-10-27 18:06 UTC
WHAT CHANGEDGitHub published CVE-2026-114612 on 2026-10-27, a high-severity privilege escalation vulnerability in LiteLLM. The flaw allows attackers to gain proxy admin access through cross-domain reuse of the salt key. Patched versions are 1.100.4, 1.101.3, 1.102.2, 1.103.1, and 1.104.0rc2.
WHY IT MATTERSAn attacker exploiting this can escalate to full proxy admin privileges, compromising authentication boundaries across domains where the same salt key is reused. Any LiteLLM proxy deployment using affected versions is exposed to unauthorized administrative access.
WHAT TO DOIdentify LiteLLM proxy deployments and their versions. Upgrade to the patched version matching your release branch: 1.100.4, 1.101.3, 1.102.2, 1.103.1, or 1.104.0rc2. Verify no shared salt keys remain across domain boundaries after patching.
Model-reported confidence 85% (not measured accuracy)
Primary source ↗
written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+ghsa_id: GHSA-7hp6-4w63-5g45 +summary: LiteLLM: Privilege Escalation to Proxy Admin via Cross-Domain Reuse of the Salt Key +severity: high +published_at: 2026-10-27T15:04:14Z +cve_id: CVE-2026-114612 +vulnerabilities[].package.name: litellm, litellm, litellm, litellm, litellm +vulnerabilities[].first_patched_version: 1.100.4, 1.101.3, 1.102.2, 1.103.1, 1.104.0rc2
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
| triage verdict | MATERIAL at 0.92, needed 0.5 |
|---|---|
| triage reasoning | LiteLLM is an AI gateway SDK in many service dependency trees. Privilege escalation to admin is actionable. |
| signals | AI provider SDK affected (LiteLLM) · privilege escalation vulnerability · high severity rating · multiple patched versions released · salt key reuse across domains — authentication/crypto weakness |
| suppression | — |
| triage model | anthropic/claude-haiku-4.5 ($0.00000) |
| writer model | anthropic/claude-sonnet-4.5 ($0.00000) |