Litellm

Litellm is a lightweight language-model abstraction library that sits in the request path between applications and multiple LLM APIs, providing a unified interface for model routing and cost management across vendors such as OpenAI, Anthropic, and others. The vendor's vulnerability profile skews toward serious outcomes: a meaningful share reach critical severity, and the disclosures have an elevated tendency toward both confirmed in-the-wild exploitation and public exploit availability. The recurring exposure centers on the single Litellm product and clusters around input-handling and authorization weaknesses—notably SQL injection, code injection, and authorization bypass—that arise from the library's role parsing user inputs, dynamically handling API requests, and mediating access to backend LLM services. Given the library's position in production AI pipelines and its bridging function between untrusted application code and high-value model APIs, defenders should treat this vendor's advisories as high-priority and audit downstream applications that integrate it; live exploitation and severity counts are shown alongside this summary.