Preserving the Freedom to Learn in AI

• Lawful access to info = lawful right to learn from it — human or machine.
  • Scale of automation doesn’t change legality vs. manual copying.
  • Courts already rejected publisher overrides on first-sale doctrine.
• Three threats: publisher AI bans, training copyright suits, inference control.
• Restrictive licensing crushes startups; only incumbents can afford deals.
  • “Little Tech” locked out while Big Tech negotiates its way in.
• robots.txt model — voluntary, machine-readable — should replace legal mandates.
• Contracts that negate copyright’s public protections should be unenforceable.
• Five policy asks: affirm learning rights, ban infringing outputs, Open Data Commons.
  • Getty/Perplexity and Time/OpenAI show licensing works for non-public data.

Matt Perault (a16z Head of AI Policy) · 2026-02-11 · Read on a16z.com