PT-2026-35643 · Undefined · Undefined
Published
2026-04-27
·
Updated
2026-04-29
·
CVE-2026-42208
None
No severity ratings or metrics are available. When they are, we'll update the corresponding info on the page.
Critical pre-auth SQL injection in LiteLLM AI gateway exploited within 36 hours of disclosure. Attackers targeted high-value tables containing provider credentials and API keys with surgical precision.
Technical details:
• CVE-2026-42208 in LiteLLM versions 1.81.16-1.83.6 allows arbitrary SELECT queries via Authorization header
• Attacker used
sk-litellm' UNION SELECT payloads targeting litellm credentials, LiteLLM VerificationToken, and litellm config tables
• Exploitation from IPs 65[.]111[.]27[.]132 and 65[.]111[.]25[.]67 using Python/3.12 aiohttp/3.9.1 user-agent
• Operator demonstrated schema knowledge, switching from lowercase to PascalCase table names when queries failedAttack methodology:
• Phase 1: Direct targeting of three highest-value tables containing OpenAI/Anthropic credentials
• Phase 2: Column-count enumeration using varying NULL placeholders (T1190, T1552.001)
• No generic SQLmap patterns - deliberate, customized exploitation suggesting prior schema analysis
• IP rotation between phases indicates sophisticated tooling to evade rate limits
DFIR opportunities:
• Web logs showing POST /chat/completions with
Authorization: Bearer sk-litellm' patterns
• Requests containing UNION SELECT, table names litellm credentials, "LiteLLM VerificationToken"
• Python aiohttp user-agent combined with SQL injection indicators
• Upstream provider billing anomalies from unexpected IP ranges#DFIR Radar
Found an issue in the description? Have something to add? Feel free to write us 👾
Related Identifiers
Affected Products
Undefined