Sysdig Reports Targeted Exploitation of LiteLLM SQL Injection Flaw

Summary
Sysdig reports targeted exploitation of LiteLLM CVE-2026-42208 began about 36 hours after disclosure, using crafted Authorization headers to query secrets. LiteLLM 1.83.7 fixes the unauthenticated SQL injection; rotate credentials or use the stated workaround.
Key points
- CVE-2026-42208 is an unauthenticated SQL injection in LiteLLM proxy API key verification, exploitable with a crafted Authorization header sent to an LLM API route.
- The flaw can allow reading or modifying the proxy database, which may contain API keys, provider credentials, and environment or configuration secrets.
- Sysdig researchers reported targeted exploitation starting about 36 hours after public disclosure on April 24, including crafted requests to /chat/completions that queried tables holding credentials and other secrets.
- Sysdig said the operator later switched IP addresses and used fewer, more precise payloads based on table names and structures identified earlier.
- LiteLLM 1.83.7 fixes the flaw by replacing string concatenation with parameterized queries.
- Researchers advised treating internet-exposed instances running vulnerable versions as potentially compromised and rotating stored virtual API keys, master keys, and provider credentials.
- For operators unable to upgrade, maintainers recommend setting disable_error_logs: true under general_settings as a temporary workaround.
Article Details
- Vulnerability Types
- SQL injection
- Severity
- Critical
- Exploitation Status
- active
- Exploit Availability
- unknown
- Patch Status
- available
- Workarounds
- If unable to upgrade, set disable_error_logs: true under general_settings.
