Text and code attribution for LLMs and coding assistants.

From book-length memorization to GitHub-license contamination. Detection, attribution, and evidence for text-based generative models.

• nv-recall
  Stanford Algorithm 1 — verbatim memorization detection
• Min-K%++
  Likelihood-based MIA (Zhang et al., ICLR 2025)
• DE-COP
  Black-box multiple-choice probing (Duarte et al.)
• Cooper PDE
  Probabilistic discoverable extraction (Cooper et al., 2024)
• CodeMIA
  Code-specific membership inference
• LLM clone detector
  Model lineage analysis
• License contamination
  GPL / MIT / Apache violation detection
• RAG two-stage analysis
  For retrieval-augmented systems
• Concept memorization
  Distributional memorization via MMD / KS

Author estate monitoring a frontier model for book memorization.
Publisher pursuing a Kadrey-style matter.
Software vendor investigating coding-assistant license contamination.
News org checking for article-level extraction.