Particle.news

Hackers Exploit Critical LiteLLM SQL Injection Within 36 Hours of Advisory

The flaw exposes databases that often store provider and cloud credentials with broad access.

Overview

  • Sysdig reported that exploitation began about 36 hours after the advisory was indexed on April 24, describing the activity as a targeted schema probe rather than a confirmed breach.
  • The bug lets an unauthenticated request with a crafted Authorization header reach a vulnerable key‑check query and read or change the LiteLLM proxy database.
  • Affected releases are versions 1.81.16 through 1.83.6, and maintainers fixed the issue in version 1.83.7 by switching to parameterized queries.
  • Researchers saw precise queries against tables that hold API keys and configuration data, along with column and schema enumeration in a two‑phase operation using rotated IP addresses.
  • Maintainers urge immediate upgrading or the disable_error_logs workaround, and operators should rotate all stored keys because rows can include high‑value OpenAI, Anthropic, and AWS Bedrock credentials.