GitHub Releases AI Universal Secret Detection Model to Strengthen Push Protection
Understand the story
On October 8, 2026, GitHub announced on its official blog the launch of an AI classifier fine-tuned from ModernBERT to extend its push protection feature. The model can evaluate candidate secrets in context, processing each batch in under 2 milliseconds; compared with the previous approach, the number of blockable secrets is expected to more than double. The feature was in private preview at the time. On October 14, GitHub published another post introducing this context-aware secret classifier: it uses a two-stage pipeline that first generates candidate secrets with regex, then uses the fine-tuned ModernBERT classifier to distinguish real credentials from similar false positives. Compared with the previous LLM-based pipeline, weighted alert precision improved by 11.5%, median post-push scan execution time dropped by 86%, and service costs were cut by more than 98%.
AI overview of reporting · Updated 5 days ago
Developments
2 developments
- 10/14 01:56 · 1 reportsGitHub Launches Context-Aware Secret ClassifierGitHub Blog · AI & ML：GitHub’s model for secret detection: Finding more secrets with less noise
- 10/8 01:45 · 1 reportsGitHub releases AI universal secret detection model to strengthen push protectionGitHub Blog · AI & ML：Secret protection must scale with software
Reporting timeline
Explore different aspects of the story through its reporting.
- GitHub Blog · AI & MLGitHub’s model for secret detection: Finding more secrets with less noise
GitHub has introduced a context-aware secret classifier that distinguishes real credentials from similar false positives via a two-stage pipeline (regex candidate generation plus fine-tuned ModernBERT classification). Compared with its previous LLM-based pipeline, weighted alert precision improved 11.5%, median post-push scan execution time dropped 86%, and service costs were cut by more than 98%.
- GitHub Blog · AI & MLSecret protection must scale with software
GitHub released a fine-tuned ModernBERT classifier for push protection that evaluates candidate secrets in context in under 2 milliseconds per batch, potentially more than doubling the number of secrets it can block; the feature is currently in private preview.
Story heat trend
Current heat3·Comparable peak10（10/14 03:00）·Comparable change over 24 hours -49%
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.