Ai Training Data Use Demand Letters

Challenges for plaintiffs:

• Black box problem: Training datasets and model architectures often not publicly disclosed
• Discovery needed: Requires litigation to compel disclosure of training data sources
• Probabilistic outputs: Hard to prove specific output is “copy” vs. coincidental similarity

Evidence plaintiffs can gather:

• Public dataset disclosures: Common Crawl, LAION, Books3 have been documented; check if your works are included
• Prompting for reproduction: Test AI with prompts designed to elicit your copyrighted work (e.g., “Write article about [topic] in style of [Your Name]”)
• Metadata analysis: Some outputs contain artifacts suggesting training on specific sources
• Company statements: Public disclosures about training data sources
• Scraped content logs: Web server logs showing AI company bots scraping your site

AI training cases are natural class actions:

• Common questions: Did defendants scrape/train on works without permission? Is it fair use?
• Large classes: Millions of creators whose works were scraped
• Settlement leverage: Class certification creates existential risk for AI companies
• Opt-out rights: Class members can opt out to pursue individual claims if they have strong damages cases

• Individual vs. collective action: Consider joining existing class actions vs. individual demand
• Publicity: Public demands/lawsuits attract media attention, putting pressure on AI companies
• Licensing opportunity: Frame as “We’re open to licensing our content for AI training at fair rates”
• Discovery needs: Litigation may be necessary to uncover what training data was used

• Firm but business-oriented: “We recognize AI’s potential but demand fair compensation for our creative works”
• Open to licensing: “We’re willing to negotiate reasonable licensing terms for training use”
• Cite precedent: Reference licensing deals AI companies have made with other publishers
• Collective strength: If representing multiple creators, emphasize scale of infringement

For News Publishers:

• Emphasize paywall circumvention and ToS violations
• Reference NYT and other publisher litigation as precedent
• Highlight existing licensing deals (OpenAI-News Corp, etc.) as proof of market value
• DMCA §1202 claims for copyright management information removal

For Individual Creators:

• Consider joining class actions rather than individual demands (cost-effective)
• If pursuing individually, focus on works with clear commercial value and registration
• Document attempts to opt out (robots.txt, no-scraping notices)

For Software Developers:

• Open-source license violations (GPL requires attribution/sharing; Copilot doesn’t comply)
• Specific code snippets reproduced in outputs
• Loss of attribution and credit

Four-factor analysis:

• No substantial similarity: Outputs don’t reproduce copyrighted expression; only rare “memorization” cases show copying
• No market harm: Plaintiffs can’t show lost sales/licenses caused by AI training
• Implied license: Publicly posting content online creates implied license for certain uses
• First sale doctrine: AI company lawfully acquired copies (e.g., purchased books) and can use for training
• Statute of limitations: Claims accrued when training occurred (3 years for copyright)

• Evaluate claim strength: Is plaintiff’s work actually in training data? Can they prove it?
• Discovery burden: Plaintiffs need litigation to compel disclosure of training data; expensive for individuals
• Settlement vs. licensing: Consider whether licensing deal is cheaper than litigation
• Collective approach: Industry-wide licensing standards emerging; individual deals may set precedent
• Policy advocacy: Support legislative solutions creating AI training exemptions or compulsory licensing

Growing trend toward negotiated licenses:

• Publisher deals: OpenAI-News Corp, Google-AP, etc. – typically $X million annually for training access
• Opt-in registries: Some platforms allow creators to register works for AI training for compensation
• Collective licensing: CMOs (collective management organizations) for music/publishing could administer AI licenses
• Statutory licensing: Possible future legislation creating compulsory licenses (like music mechanical licenses)

• Evaluate whether your works were used in AI training (dataset analysis, output testing)
• Draft demand letters and licensing proposals to AI companies
• Negotiate licensing agreements for authorized AI training use
• Join or initiate class action litigation
• File individual copyright infringement lawsuits when damages justify
• Pursue DMCA §1202 claims for CMI removal
• Assert ToS violations and breach of contract claims

• Assess fair use and other defenses to training-based infringement claims
• Respond to demand letters and evaluate settlement vs. litigation
• Negotiate licensing agreements with content owners
• Defend copyright infringement and class action lawsuits
• Advise on training data sourcing and documentation
• Implement opt-out mechanisms and respect robots.txt / no-scraping signals
• Develop industry-standard licensing frameworks

• News publisher claims against LLM developers
• Book author and artist class actions
• Software developer Copilot disputes
• Licensing negotiations for AI training data
• Fair use defense in AI training cases
• DMCA §1202 CMI removal claims
• Privacy-based claims (CCPA, BIPA) for facial recognition / personal data training

Book a call to discuss your AI training dispute. I’ll assess the strength of infringement or fair use claims, evaluate litigation vs. licensing options, and recommend strategy for resolution or defense.

Email: owner@terms.law