Fairly Trained is a UK-based organization, founded in 2024 by Ed Newton-Rex, that certifies AI developers whose models are trained exclusively on licensed or consented data. It operates as an independent certifier of training-data provenance, providing a market signal intended to distinguish developers that license their training data from those that rely on fair use claims.
Overview
Fairly Trained was founded in 2024 by Ed Newton-Rex, the former Vice President of Audio at Stability AI. Newton-Rex resigned from Stability AI over disagreements about the company's approach to copyrighted training data, and Fairly Trained was established as a response to the copyright controversies surrounding the generative AI industry. The organization's certification covers AI developers that train their models exclusively on licensed or consented data, creating a market signal for what it describes as ethically sourced training data.
Certification model
Fairly Trained certifies training-data provenance as an independent body, positioning its certification as a market-based mechanism addressing the AI Copyright disputes that have driven litigation including NYT v. Microsoft, OpenAI et al.. By creating a visible distinction between companies that license training data and those that rely on fair use claims, the certification is intended to allow rights holders and consumers to make informed choices. The approach is presented as an alternative to purely regulatory solutions.
The organization has certified several AI companies, including ElevenLabs for certain product lines.
Critics question whether voluntary certification can meaningfully shift industry practices while the largest frontier labs continue to train on web-scraped data under fair use arguments.
Relationships
- supports: AI Copyright — provides market-based mechanism for training data consent
- related: NYT v. Microsoft, OpenAI et al. — certification model responds to the same copyright tensions driving litigation
- related: ElevenLabs — among companies engaging with Fairly Trained certification
- contradicts: Fair use arguments advanced by major frontier labs for web-scraped training data