AI Stack Alert High For CISO, SOC / IR, AppSec, AI Eng AI-assisted
ZDI-26-1031: LiteLLM MCP stdio command injection allows authenticated remote code execution
Zero Day Initiative published advisory ZDI-26-1031 for LiteLLM. The flaw is in handling of MCP stdio configurations, where a user-supplied string is used in a system call without proper validation. An authenticated remote attacker can execute code in the context of the service account.
Why it matters
Organizations running LiteLLM with MCP stdio enabled expose their service account to code execution by anyone holding valid credentials. The vendor disputes that the report describes a vulnerability and says no separate patch is coming.
Do this
- Upgrade LiteLLM to 1.104.0 or later and leave LITELLM_ENABLE_MCP_STDIO unset so stdio MCP servers stay disabled.
- Replace any default or example master key with a strong random value; v1.104.0+ refuses startup if the key is unset, empty, or sk-1234.
- Restrict who can create or test MCP server configurations to trusted proxy administrators.
- Do not rely on the CVE-2026-42271 allowlist as a security control, since the vendor states it is not one and will be removed.
- Review existing MCP stdio server configurations in LiteLLM for unexpected host commands.
Signals
- NoCISA KEV
- 35/100Priority
- 75%Confidence
Evidence
“The specific flaw exists within the handling of MCP stdio configurations.”
“The issue results from the lack of proper validation of a user-supplied string before using it to execute a system call.”
“From v1.104.0, stdio servers are disabled unless LITELLM_ENABLE_MCP_STDIO=true is set in env”
Sources
Affected / related
LiteLLM, OS command injection, Zero Day Initiative
Published 2026-11-18 06:00 UTC · updated 2026-11-18 20:35 UTC · AI-assisted summary, checked against quoted sources · How we score