Does using AI code assistants involve OSS plagiarism risk?
summary
Study scans 10,000 LLM-generated snippets against a ~16 TB OSS knowledge base covering ~250 million projects and finds 30% of AI snippets contain ≥10% OSS, versus prior 0.8–2.2% results on much smaller corpora. Matching uses hash-based snippet matching to detect small copied fragments, not just dependency manifests. Conclusion: treat LLM output as potential OSS, add automatic snippet-level SCA in CI/CD and verify licenses before merging.
AI-written summary · about
chapters (8)
- 0:00Introduction and agenda
- 1:05AI code assistants risks explained
- 2:30Legacy software composition analysis
- 3:36Software Transparency Foundation
- 5:01Research methodology details
- 6:25Study results and findings
- 7:28Key takeaways and call to action
- 8:30Q&A session start
related talks
24mLicense compliance with AI assisted coding
38mOpen Source Licensing in the Era of AI-Assisted Coding
29mAI-Generated Code: Legal Risks and How to Reduce Them
38mContributing AI-Generated Content to OSS Projects - Navigating Legal, Copyright, &...
48mOpen Source: What's Next
31mUsing Generative AI to Develop Open Source Code - Practical Guidance for Managing Lega...
38mWhat Should OSPOs Think About Github CoPilot?
38mThe Cognitive Cost of AI-Assisted Open Source