Your Training Data Has a Lawyer Problem (And Your Lawyer Has a Data Problem)

Your Training Data Has a Lawyer Problem (And Your Lawyer Has a Data Science Problem)

Abstract:

Every data scientist has assembled a dataset by scraping, licensing, purchasing, or quietly reusing something that was already sitting in the warehouse. Most of the time nobody asks where it came from — until a customer, a regulator, or opposing counsel does.

This talk is a product lawyer’s field guide to data provenance, delivered for people who actually build models. We’ll walk through the questions Legal asks and *why* — what makes a dataset licensable vs. merely available, why “it’s public” is the most expensive sentence in AI, how scraped data, user-generated content, and vendor-supplied corpora carry fundamentally different risk profiles, and what a data scientist can do at collection time that saves a year of pain at deployment time. We’ll cover real patterns from AI supplier reviews and contract negotiations — the retention clauses, the “we may use your inputs to improve our service