GitHub Blog · AI & ML · Batu Ozturkler · 20 天前模型评分49
GitHub 新密钥检测模型：用更少误报发现更多密钥
GitHub’s model for secret detection: Finding more secrets with less noise
导读
GitHub 为密钥扫描训练了专用分类器，采用正则筛选加 ModernBERT 上下文分类的两阶段流程，替代此前的 大语言模型（LLM，Large Language Model） 流水线。相比旧方案，用户告警接受率提升 11.5%，推送后扫描中位执行时间缩短 86%，服务成本降低 98% 以上。该效率使通用密钥检测得以扩展到推送保护，在代码推送前就返回结果。
分类置信度
高
- 官方或准官方来源，且无冲突类别。
- 正文、可信日期、主分类和证据摘录完整。
来源级别：官方一手
事件类型：产品更新
证据摘录
- “To address this challenge, we trained a secret classifier specifically for GitHub’s product needs: distinguishing real credentials from lookalikes while meeting the speed and efficiency requirements of scanning at scale.”
- “We developed a fast classifier that reduced median post-push scan execution time by 86% compared with the previous LLM-based pipeline.”
- “That efficiency made it practical to extend generic secret detection to push protection, returning results before developers push changes, not only after code is committed.”
- “The trained model is exported to ONNX format with 8-bit quantization, enabling fast inference.”
来源：GitHub Blog · AI & ML · github.blog