← AI security in plain English

LLM03

OWASP Top 10 for LLMs / Supply Chain

Weaknesses enter through the outside pieces an LLM depends on, such as third-party models, training datasets, and software libraries, any of which can be tampered with or poorly vetted.

Think of it likeBuilding your house with lumber and locks from a stranger's truck, never checking whether any of it is rotten or rigged.

In plain English

AI systems are assembled from borrowed models, data, and code; if any of those outside parts is tampered with or flawed, the whole system inherits the problem.