Definition
An attack that targets the software packages, libraries, datasets, or tools that developers use to build AI systems, rather than attacking the finished AI product directly. By poisoning a widely used building block — such as an open-source AI library — attackers can compromise every product built on top of it.
Why it matters
AI developers routinely install hundreds of open-source packages from public repositories. A single compromised package can give attackers a backdoor into AI coding tools, developer workstations, and production systems across many organisations simultaneously.