A license grants permission to use protected material under specified conditions. For GenAI, verify rights across the vendor’s training corpus, your fine-tuning data, and retrieval documents before they enter a pipeline. section 6.6.1
Public availability does not establish permission for commercial redistribution. Once unlicensed material enters training, fine-tuning, or retrieval, it can propagate into model weights, embedding stores, and derivative systems; removing it may require costly, sometimes impossible, retraining. section 6.6.1
ClaimAssist’s unlicensed articles
A vehicle-claims model reproduced automotive website advice nearly word for word, although the articles were not licensed for commercial redistribution. The company withdrew the model, rebuilt its dataset, and reran evaluations, taking weeks and incurring costs beyond upfront licensing. section 6.6
Figure 6.12 IP risk lifecycle in GenAI deployments. Risk enters before ingestion through training data, fine-tuning datasets, and retrieval documents; propagates during use as memorized content and contractual terms shape what the model retains; and surfaces after release as verbatim reproduction, exfiltration, or model distillation. Arrows show how a licensing failure at ingestion can reappear as an output infringement claim months later.

Trace how licensing risk can enter through data, persist during use, and surface after release as reproduced content or other claims. section 6.6



