AI supply chain attack

An AI supply chain attack goes after the pieces an AI system is built from, not the finished tool. Those pieces include models, datasets, code libraries, plug-ins, containers and the services that host them. Think of a bakery that buys flour from a supplier. If someone tampers with the flour, every loaf is affected, even though nobody broke into the bakery. In AI, the tampered flour could be a model file shared on a public hub, a harmful package named almost like a popular library, or a plug-in that has been compromised.

Why it matters

Your organization may never write code, but the AI tools you use still depend on parts made by others. A tampered part can cause harm you cannot see from the outside, sometimes through a backdoor hidden in a model. Security groups treat this as a top risk. The OWASP Top 10 for LLM Applications 2025 lists supply chain problems, and MITRE ATLAS lists AI supply chain compromise. Install AI software and models only from official sources, and keep them updated. When you choose a vendor, ask where their models and plug-ins come from and how they check them. We publish known problems on our advisories page.