The AI technology stack is the full environment that supports an AI system, not just the model itself. It includes the hardware that runs inference and training, cloud services, networking, data pipelines, labeling and storage systems, and the application layer that users interact with.
This matters because many AI security failures happen outside the model weights. Attackers may target weak access controls, poisoned data feeds, insecure APIs, compromised cloud credentials, or the deployment environment to cause leakage, denial of service, or unsafe outputs. Defenders therefore need to secure the entire stack with inventory, least-privilege access, monitoring, and testing of each layer. In practice, red-teaming and prompt-injection testing often focus on the application layer, while supply-chain and data-integrity checks protect the upstream pipeline. Treating the stack as one system helps security teams see how model risk, infrastructure risk, and data risk connect.