Check Point Research disclosed two critical vulnerabilities in LiteLLM, a gateway handling 97 million API calls monthly. CVE-2026-35029 lets any authenticated user reach admin functions, executing arbitrary code via malicious Python modules. CVE-2026-59823 redirects gateway traffic to steal LLM provider credentials.
The core flaw: read-only users escalate to admin. LiteLLM’s model doesn’t distinguish privilege levels, only API key existence. In shared deployments, that’s catastrophic.
A gateway sits between your application and expensive LLM providers. Compromise it and attackers control the channel, stealing credentials and exposing every downstream system.
Researchers demonstrated complete system compromise: file access, credential theft, remote code execution. Both flaws are patched in v1.83.9+. Disclosure began October 2025 with active exploitation already reported. If you run LiteLLM, upgrade now. If you’re evaluating gateways, auth design matters.
Sources: When LLM Gateways Go Wrong: Findings from LiteLLM • Check Point Research Discovers High Risk Vulnerabilities in LiteLLM
Coverage: CSA Research Note: LiteLLM RCE via MCP Injection • SEC Consult: Broken Access Control in Config Endpoint
Related on this blog: [AUTO] MemTensor Worm Exploits GitHub Actions Token Interception • [AUTO] Encrypted Instructions Trick Copilot CLI Into Stealing Secrets • FortiBleed Ran on a Command Fortinet Ships in the Box