AI Model Versioning and Rollback Strategy

AI model providers update their models periodically, sometimes changing behaviour in subtle ways that can affect a production feature built around a specific model version.

Pinning to a specific model version, testing new versions against a fixed set of evaluation cases before upgrading, and keeping a rollback path ready are all standard software practices that apply directly to AI models.

We treat model upgrades as a deliberate, tested release rather than something that happens automatically in the background.