Integrity validation is the process of confirming that model weights, training data, prompts, dependencies, and configuration files match an expected trusted state. It usually relies on checksums, hashes, digital signatures, or verified baselines to detect changes that were not approved. In AI systems, this matters because a small alteration can change behavior without breaking the software.
Defenders use integrity validation to spot tampering, supply-chain compromise, poisoned data, or accidental configuration drift before the system is trusted again. It is especially important during recovery: bringing an AI service back online is not enough if the model or its inputs were silently modified. In attacks, adversaries may replace files, alter prompts, or change API settings to steer outputs. Integrity validation helps prove whether the system is still the one you intended to run.