LiteLLM

LiteLLM works like a universal power adapter for AI models made by different companies, letting developers swap one model for another without rewriting much code. If you're using an OpenAI model and want to switch to a Google model, you'd normally have to learn a whole new way of calling it — but with LiteLLM in between, you keep the same interface and just switch out the model behind it.

As startups and large companies rushed to add AI features, this kind of integration layer became widely adopted. The catch is that LiteLLM is distributed through a Python package repository, so if that distribution channel itself gets attacked, every organization using the tool can be exposed at once. In fact, a 2025 supply chain attack saw a compromised version of the tool circulated, leading to a large-scale credential leak.

The term shows up in two different kinds of articles. One is incident coverage, where a compromised version was distributed and identified as the path through which access keys from over 2,500 organizations, including Microsoft and Amazon, were leaked. The other is routine usage coverage, where n8n's new agent node listed the tool as one of its supported providers. One common misunderstanding worth noting — the tool itself is not malware. A normally distributed version was tampered with by attackers; the tool's core function isn't designed to exfiltrate data.