5/7/26 LiteLLM is an AI proxy, a component of today’s AI infrastructure, that rcently had a critical vulnerability which was exploited within 36 hours of public indexing, underscoring how quickly threat actors operationalize AI infrastructure weaknesses. The flaw enabled pre-auth SQL injection attacks against LiteLLM proxies, exposing sensitive assets such as API keys, provider credentials, and environment configurations stored in backend databases.
For security leaders, this is a reminder that AI gateways and orchestration layers are now part of the critical attack surface. Rapid patching, reducing exposed services, hardening error handling, and protecting secrets management are essential controls as attackers increasingly target the AI software supply chain.
Notable observations were that attackers systematically tested the database structure to craft precise extraction queries rather than blindly attacking the system. This shows that attackers were knowledgeable about the application architecture and indicates that threat actors are moving from opportunistic attacks to highly targeted AI platform exploitation. The speed and sophistication of the reconnaissance phase was significant, even though there was no confirmed follow-on abuse of the stolen credentials.
https://www.securityweek.com/fresh-litellm-vulnerability-exploited-shortly-after-disclosure/