Model deprecation is the planned retirement of an AI model version. After deprecation, the old model may stop accepting requests, require a new endpoint, or return compatibility errors instead of behaving as it did before. This is different from an outage: deprecation is intentional and usually announced, while an outage is an unplanned service failure.
In cyber security, deprecation matters because applications, automations, and agents can break when they depend on a specific model ID, prompt format, or safety behavior. Attackers may exploit weak migration planning by targeting systems that have not been updated to newer APIs or guardrails. Defenders use deprecation to remove vulnerable, inaccurate, or costly models, but they need version inventories, fallback logic, and testing so security tooling and business workflows keep working. Good change management treats model retirement like any other infrastructure change: document it, monitor failures, and provide a controlled migration path.