OpenAI 2019 LibGen disclosure to Gates preceded 2024 Authors Guild evidence of internal legality concerns
Authors Guild filings establish executive knowledge of LibGen piracy for GPT training from 2019 onward. Internal assessments weighed optics over compliance and projected writer displacement. Evidence supports expanded discovery into dataset scale and downstream licensing negotiations.
Newly unsealed briefs detail OpenAI and Microsoft awareness of LibGen sourcing as early as 2019. Internal messages reference the dataset as sketchy while proceeding with ingestion of millions of copyrighted volumes. Policy staff tracked downstream effects on genre authors and employment displacement yet classified complaints as acceptable disruption.
Data from the filings cross-reference statements by Jack Clark in 2020 on GPT models substituting for Amazon writers and Tarun Gogineni memos on completing George R.R. Martin series. Microsoft received contemporaneous notice yet maintained partnership. Parallel media-plaintiff dockets record similar dataset provenance tracking.
These records connect to prior copyright actions including New York Times v. Microsoft and ongoing EU AI Act data provenance rules. They establish a pattern of risk acceptance across training runs exceeding 300 billion tokens from unauthorized sources.
Operational outcome centers on liability exposure for downstream model releases. Regulators now possess timestamped internal acknowledgments that can anchor enforcement on training data acquisition practices through 2026.
District Court: denial of OpenAI summary judgment motion on willfulness by Q3 2025
Sources (3)
- [1]Authors Guild v. OpenAI Plaintiffs Brief(https://authorsguild.org/news/ag-v-openai-top-execs-knew-mass-book-piracy-was-illegal/)
- [2]New York Times Co. v. Microsoft Corp. Docket(https://www.courtlistener.com/docket/67810529/)
- [3]EU AI Act Recital 105 Training Data Transparency(https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689)