GitHub AI secret detection for push protection

GitHub says a new ModernBERT-based classifier can recognize unstructured secrets in code context and help block them before a push enters repository history. The push-protection feature is in private preview, with availability for eligible GitHub Secret Protection organizations planned later in October. GitHub says existing post-push AI secret detection is being updated now. It reports that its internal measurements found 2.84 times as many screened public pushes between the second quarters of 2024 and 2026, without a statistically detectable rise in the share containing credentials. Those are GitHub’s own data and performance claims; the future impact of the classifier remains unproven.