Fair use argument
Training as transformative. Doesn't reproduce originals. Similar to Google Books case. Uncertain in US courts.
Advertisement
Reproduction claim
Plaintiffs show verbatim reproduction from training data. NYT article verbatim from GPT-4. Damages claim.
Advertisement
Opt-out regimes
UK TDM exception with opt-out. EU CDSM Article 4 with opt-out via metadata. HTTP header/robots.txt for AI crawlers.
Licensed data
Trend: OpenAI licenses NYT, Reuters, AP. Anthropic licenses various. Move away from unlicensed scraping for high-value data.