AI Threat Watch

Case litellm-cve-2026-42271Agent hijackWorldwide

CISA adds actively exploited LiteLLM AI gateway RCE (CVE-2026-42271) to KEV

Attackers exploited command injection in LiteLLM's MCP server preview endpoints, chainable with a Starlette Host bypass for unauthenticated RCE and theft of AI provider keys.

CriticalAttack on AIStatus: PatchedVerified case file

Open in the interactive mapCite this case

Reported
Jun 8, 2026
Location
Worldwide
Target
BerriAI LiteLLM proxy
Severity
Critical
Triggered by
Unknown attackers exploiting in the wild
Actor type
Unknown
Reported loss
None reported
CVE
CVE-2026-42271, CVE-2026-48710

What happened

Attackers exploited command injection in LiteLLM's MCP server preview endpoints, chainable with a Starlette Host bypass for unauthenticated RCE and theft of AI provider keys.

Impact: Unauthenticated remote code execution on exposed LiteLLM proxies and theft of stored AI provider keys; number of compromised deployments not disclosed

Who triggered it

Actor: Unknown attackers exploiting in the wild

Attribution: No public attribution; CISA confirmed active exploitation via KEV listing; chain validated by Horizon3.ai

Motive: Takeover of AI gateways and theft of upstream AI provider API keys

How it happened

Attack vector: POST to /mcp-rest/test/connection or /mcp-rest/test/tools/list with a stdio MCP config whose command is spawned as a subprocess; chained with Starlette BadHost (CVE-2026-48710) to skip auth

Root cause: MCP preview endpoints executed user-supplied commands and were exposed to any API key holder; Starlette rebuilt request paths from unvalidated Host headers, letting auth middleware be bypassed

Timeline

  1. CVE-2026-42271 disclosed in LiteLLM 1.74.2 through 1.83.6
  2. LiteLLM 1.83.7 restricts MCP test endpoints to PROXY_ADMIN
  3. Starlette CVE-2026-48710 BadHost disclosed; Horizon3.ai validates unauthenticated chain June 1
  4. CISA adds CVE-2026-42271 to KEV with federal deadline June 22

Response

BerriAI shipped 1.83.7, Starlette 1.0.1 fixed BadHost; CISA KEV listing with remediation deadline; WAF blocking of the test endpoints advised

Current status: Patched

Sources (open in a new tab)

Cite this case

CISA adds actively exploited LiteLLM AI gateway RCE (CVE-2026-42271) to KEV. AI Threat Watch, case litellm-cve-2026-42271, Jun 8, 2026. https://aithreat.io/cases/litellm-cve-2026-42271/

Permanent link: https://aithreat.io/cases/litellm-cve-2026-42271/

This case file in other languages: EnglishFrançaisEspañolالعربية中文