Tools AI Risk Radar ai-incident-0105
Incident record
LiteLLM let signed-in users send the proxy's API keys elsewhere
- Severity
- Medium
- Status
- Patched
- Type
- Data Leak
- Target
- LiteLLM proxy (BerriAI) before 1.88.6 and 1.96.2
- Actor
- researcher
- CVE
- CVE-2026-84377
What happened
Any authenticated user of a LiteLLM proxy before 1.88.6 and 1.96.2 could point an outbound model call at a server they controlled, and the proxy sent its configured provider credentials there. LiteLLM's request validation missed routing and credential parameters such as api_base, base_url, model_list, fallbacks and litellm_credential_name when they arrived in nested fields, path values or bracket-notation form data. The CVE record of 2 September 2026 scores it CVSS 3.1 6.5.
LiteLLM's advisory GHSA-3cv6-jpf6-8222 names 1.88.6 and 1.96.2 as the fixed releases. A separate LiteLLM flaw published on 28 September, CVE-2026-93355, lets one login token take over another user's account and is listed on its own row.
Sources
- NVD record CVE-2026-84377nvd.nist.gov/vuln/detail/CVE-2026-84377
- LiteLLM advisory GHSA-3cv6-jpf6-8222github.com/BerriAI/litellm/security/advisories/GHSA-3cv6-jpf…
One record from the AI Risk Radar, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 13 October 2026