800 artists call AI training "theft." Here's what it means for your creative business
Hundreds of artists-among them Scarlett Johansson, Cate Blanchett, R.E.M., and Breaking Bad creator Vince Gilligan-have signed an open letter from the Human Artistry Campaign's "Stealing Isn't Innovation" effort. Their message: stop using copyrighted work to train AI without permission and build ethical partnerships instead.
Why it matters: there are roughly 60 active lawsuits in the US against AI companies over training data, with more cases moving in Europe. Outcomes so far are mixed, and the core tension remains the same-companies argue fair use; artists say it's unauthorised copying that cuts into livelihoods and ownership.
The flashpoint
AI training pulls in massive datasets-text, images, music, video-to learn patterns and generate new content. A lot of that material is scraped from the open web without consent from rightsholders.
The debate isn't abstract. In 2024, Scarlett Johansson challenged OpenAI after an Advanced Voice sounded similar to her portrayal in "Her." Legal letters followed, and the company paused the "Sky" voice. Voice, likeness, and identity are now front-line issues, not edge cases.
The legal backdrop (in plain English)
Copyright protects original expression. AI companies claim training qualifies as fair use in some contexts; creators say wholesale ingestion of copyrighted works is unauthorised and exploitative. Courts are still sorting it out, case by case.
If you want the baseline on fair use factors, the U.S. Copyright Office has a concise overview: copyright.gov/fair-use.
What creators can do right now
- Register your work. If you're in the U.S., registration strengthens your position for statutory damages and attorney's fees. Do it as a habit, not an afterthought.
- Lock down contracts and licenses. Add "no AI training/text-and-data mining" clauses. For voice and likeness, require explicit consent, narrow usage terms, audit rights, and kill-switch provisions.
- Update your site policy and signals. Use robots.txt and relevant meta tags some crawlers respect to state "no AI training." It won't block bad actors, but it sets terms and helps with enforcement.
- Embed provenance. Add Content Credentials (C2PA) so your files carry tamper-evident context for authorship and edits: c2pa.org.
- Set up monitoring. Run periodic reverse image/audio searches, track suspicious accounts, and keep takedown templates ready (DMCA notices, demand letters, dataset deletion requests).
- Negotiate better deals. If you collaborate with AI vendors, ask for dataset transparency, opt-in only, revenue share, indemnities, and audit logs. No transparency, no deal.
- Protect your voice. Use written approvals for cloning, time-bound licenses, and strict prohibitions on model re-use or re-training. Keep a record of all voice sessions.
- Join collective efforts. Unions, guilds, and campaigns pool leverage and legal resources. A unified ask-consent, credit, compensation-moves policy faster.
- Adopt ethical AI on your terms. Use tools that license data properly and make opt-in/opt-out clear. If you're building skills, here's a curated place to start: Complete AI Training: Courses by job.
What to watch next
Expect more rulings on whether ingesting copyrighted works for training is fair use, and pressure on companies to disclose datasets. Keep an eye on opt-out registries, model provenance standards, and collective bargaining outcomes for performers and authors.
The direction is clear. Creators want consent, credit, and compensation-before their work trains models, not after. Set your policies now, tighten your contracts, and pick partners who respect your rights. That's how you keep your art-and your business-intact.
Your membership also unlocks:
