AI-generated code detection in CI/CD — looking for approaches and real-world experience [D]

Excerpt

​ I'm working on a system to estimate whether code committed to a repository was generated with AI coding tools. My current approach is based on Git/commit-level signals such as AI-related commit trailers, commit metadata, LOC changes, number of files changed, addition/deletion patterns, etc. The problem I'm running into is confidence and calibration. For example, a commit containing 500+ new lines isn't necessarily AI-generated. A developer can also modify or remove the metadata that would make