LiteLLM Flaw CVE-2026-42271 Exploited in the Wild, Chains to Unauthenticated RCE
A critical command injection vulnerability (CVE-2026-42271) in BerriAI’s LiteLLM has been actively exploited in the wild. The flaw, combined with a separate Starlette vulnerability (CVE-2026-48710), allows for unauthenticated remote code execution. Organizations using vulnerable versions of LiteLLM are advised to update immediately to mitigate the risk of unauthorized access and potential data compromise.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has classified the CVE-2026-42271 vulnerability as high severity, citing evidence of ongoing exploitation. The vulnerability resides in the LiteLLM Python package, specifically within the handling of server configurations via the `/mcp-rest/test/connection` and `/mcp-rest/test/tools/list` endpoints. Attackers can leverage this command injection flaw to execute arbitrary commands on the host system, potentially gaining complete control. Horizon3.ai has demonstrated a chained exploit combining this vulnerability with CVE-2026-48710, a host header validation bypass in Starlette, resulting in unauthenticated remote code execution. This chain has a combined CVSS score of 10.0, making it a critical threat.