The development
What happened
GitHub outlined a ModernBERT classifier developed with Microsoft's Applied Sciences group to detect unstructured credentials. Existing AI detection customers receive the model; its push-time use has a separate rollout and billing scope.
Key facts
GitHub reports candidate batches assessed in under two milliseconds; this measures model processing, not total push latency.
Push-time use remains in private preview, with later-October access planned for Secret Protection customers on Enterprise Cloud and GitHub Team.
The planned push feature consumes AI credits. Updated post-push alerts remain included with existing secret-scanning purchases.
Organizations already using AI secret detection receive the new model from 7 October; Enterprise Server and Copilot integrations are planned separately.
The context
Why it matters
Context-aware detection targets credentials that token-pattern checks can miss. Teams should distinguish the model update from planned push-time protection and check eligibility and credit use. GitHub's latency claim concerns classifier batches; it does not guarantee a two-millisecond end-to-end check or detection of every secret.