CVE-2026-42208: Targeted SQL injection against LiteLLM’s authentication path discovered 36 hours following vulnerability disclosure

CVE-2026-42208: Targeted SQL injection against LiteLLM’s authentication path discovered 36 hours following vulnerability disclosure
Sysdig TRT observed a rapid, pre-auth SQL injection (CVE-2026-42208 / GHSA-r75f-5x8p-qvmc) against LiteLLM that used UNION-based payloads to enumerate tables holding virtual API keys, provider credentials, and environment variables. The operator executed schema-aware column-count discovery from rotating egress IPs but Sysdig saw no confirmed authenticated follow-on; patch to v1.83.7 and rotate exposed keys immediately. #LiteLLM #CVE-2026-42208

Keypoints

  • Critical pre-auth SQL injection (CVE-2026-42208 / GHSA-r75f-5x8p-qvmc) in LiteLLM allowed arbitrary SELECTs against the PostgreSQL backend by concatenating the Authorization: Bearer value into a query without parameterization.
  • The advisory was indexed in the GitHub Advisory Database on 2026-04-24 and the first observed exploitation attempt occurred 36 hours and seven minutes later (2026-04-26 04:24 UTC).
  • Sysdig TRT captured precise, non-generic UNION-based schema enumeration targeting three high-value tables: LiteLLM_VerificationToken, litellm_credentials, and litellm_config (environment_variables).
  • The operator demonstrated schema knowledge (Prisma PascalCase table names) and ran textbook column-count discovery, implying prior access to the schema or deliberate reconnaissance/LLM assistance.
  • Exploitable instances expose cloud-grade provider keys and master/virtual API keys; successful extraction would allow unauthenticated reuse (replay to /chat/completions) and large blast radius across upstream accounts.
  • Recommended mitigations: update to v1.83.7, rotate all keys and provider credentials for internet-reachable instances, restrict proxies behind authenticated reverse proxies, and monitor logs for the distinctive request signatures.

MITRE Techniques

  • [T1190 ] Exploit Public-Facing Application – The vulnerability is a pre-auth SQL injection in the proxy auth-check allowing arbitrary SELECTs: β€˜the injection is fully pre-auth: any HTTP client that can reach the proxy port is sufficient.’
  • [T1567 ] Exfiltration Over Web Service – Data was returned directly in HTTP responses using UNION SELECT payloads to leak columns: β€˜that column carries the leaked data back in the response body.’
  • [T1078 ] Valid Accounts – Exfiltrated virtual or master keys could be replayed to upstream endpoints to gain access: β€˜Once a virtual key or master key is exfiltrated, it can be replayed against /chat/completions from any IP.’

Indicators of Compromise

  • [IP address ] Exploitation sources – 65.111.27.132, 65.111.25.67 (adjacent /22 in AS200373 observed issuing UNION-based SQLi and probes)
  • [ASN / Operator ] Egress attribution – AS200373 (3xK Tech GmbH) – both observed IPs belong to this operator
  • [HTTP endpoints ] Targeted URIs – /chat/completions (POST), /key/generate, /key/info (unauthenticated probes after SQLi attempts)
  • [HTTP header pattern ] Authorization header injection pattern – Authorization: Bearer header beginning with β€œsk-litellm'” (single-quote terminator) and containing SQL keywords/UNION SELECT payloads
  • [User-Agent ] Request signature – Python/3.12 aiohttp/3.9.1 (consistent across observed requests)
  • [Vulnerability identifiers ] Advisory references – GHSA-r75f-5x8p-qvmc, CVE-2026-42208 (patch available in v1.83.7)


Read more: https://www.sysdig.com/blog/cve-2026-42208-targeted-sql-injection-against-litellms-authentication-path-discovered-36-hours-following-vulnerability-disclosure