CVE-2026-114612: LiteLLM vulnerable to privilege escalation via cryptographic context reuse

# CVE-2026-114612: LiteLLM vulnerable to privilege escalation via cryptographic context reuse

https://gcdradar.vercel.app/incidents/cve-2026-114612-litellm-vulnerable-to-privilege-escalation-v · AIR-2026-1470 · closed 2026-10-22

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Patched
- First reported: 2026-10-22
- AI role: Attack on AI
- Products: LiteLLM
- CVE ids: CVE-2026-114612
- CVSS: 9.9

## Summary

A vulnerability in LiteLLM proxy server allows authenticated internal users to forge administrator credentials through cryptographic context reuse. Exploiting this flaw permits administrative access and arbitrary command execution.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Affected and fixed versions

- Affected: litellm >= 1.91.0, < 1.100.4
- Affected: litellm >= 1.101.0, < 1.101.3
- Affected: litellm >= 1.102.0, < 1.102.2
- Affected: litellm >= 1.103.0, < 1.103.1
- Affected: litellm >= 1.104.0rc1, < 1.104.0rc2
- Fixed in: litellm 1.100.4
- Fixed in: litellm 1.101.3
- Fixed in: litellm 1.102.2
- Fixed in: litellm 1.103.1
- Fixed in: litellm 1.104.0rc2

## Latest updates

- 2026-10-27: GitHub advisory published (https://github.com/advisories/GHSA-7hp6-4w63-5g45)
- 2026-10-22: CVE-2026-114612: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. (https://nvd.nist.gov/vuln/detail/CVE-2026-114612)

## Sources

- 2026-10-22 NVD: CVE-2026-114612: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. https://nvd.nist.gov/vuln/detail/CVE-2026-114612
- 2026-10-27 GitHub Advisory Database: CVE-2026-114612: LiteLLM: Privilege Escalation to Proxy Admin via Cross-Domain Reuse of the Salt Key https://github.com/advisories/GHSA-7hp6-4w63-5g45

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.