ModernBERT (AI model): news timeline & CVEs

GitHub AI Can Spot Passwords Hidden in Code Before Developers Push Them

GitHub is previewing a ModernBERT classifier that blocks hidden passwords during pushes, beyond pattern-based scanning.

GitHub and Microsoft Applied Sciences introduced a ModernBERT classifier that uses surrounding code to detect likely passwords before a push, extending protection beyond known token patterns. GitHub says candidate batches are checked in under two milliseconds and that the model could more than double secrets blocked. From Q2 2024 through Q2 2026, screened public pushes rose 2.84 times, the share containing secrets showed no clear trend, and push-protection overrides fell from 6.63% to 3.93%. The capability is in private preview for Secret Protection customers on Enterprise Cloud and GitHub Team and will consume AI credits.

Cyber Security News14:01 UTC · 8h agoAI tools & infra 2 sources

GitHub adds AI to catch passwords before a code push

GitHub is adding a ModernBERT classifier to push protection to block unstructured passwords before they enter repositories.

GitHub and Microsoft Applied Sciences built a ModernBERT classifier that expands push protection beyond recognizable credential formats to unstructured secrets such as database passwords. GitHub says the model scores batches in under two milliseconds and could more than double the secrets push protection blocks. From Q2 2024 to Q2 2026, screened public pushes rose 2.84 times and credential-bearing pushes rose 2.59 times, while about 30 percent of newly detected secrets are blocked before they enter history. The capability is in private preview, with a later-October rollout for GitHub Secret Protection on Enterprise Cloud and Team plans, and it will ship in GitHub Enterprise Server 3.23.

Help Net Securityupdated 14:01 UTC · 8h agofirst 09:02 UTC · 13h agoTools 2 sources1

Contrastive Learning for Authorship Verification

A ModernBERT bi-encoder using contrastive learning reaches 98.4% accuracy on PAN21 authorship verification.

Researchers report that contrastive learning outperforms a classification baseline for authorship verification. They identify loss function, batch size, training duration, pretrained model, input context length, and random text-span augmentation as important performance factors. Their ModernBERT bi-encoder reaches 98.4% accuracy on the PAN21 authorship verification task.

arXiv cs.AI / cs.LG / cs.CLSep 23, 17:59 UTC · 15d agoAI research

ModernBERT

Timeline3 articles

Appears with

GitHub AI Can Spot Passwords Hidden in Code Before Developers Push Them

GitHub adds AI to catch passwords before a code push

Contrastive Learning for Authorship Verification