(Preview) Apple Searches for an AI Partner, A Second Fair Use Ruling

Summary

• Andrew Sharp now leans toward treating LLM training as fair use, particularly where the resulting output is not itself infringing and the alleged harm is speculative future market harm. He still sees an uncomfortable “transfer of wealth and long-term profit incentives” from millions of creators to an industry dominated by companies worth roughly $5 trillion, but copyright cannot consistently address that harm without evidence of present market injury.
• Meta’s victory does not establish that training Llama on copyrighted books is categorically lawful. Judge Vince Chhabria ruled that these plaintiffs made the wrong arguments and lacked a record of lost market opportunities; Andrew praised that as an “intellectually honest” reckoning with both legal limits and unprecedented technological scale.
• Ben Thompson thinks Chhabria’s suggestion that stronger plaintiffs might win is “a fantasy.” Proving harm would require identifying individual purchases never made and aggregating them at scale; the harm might be real, but under the existing evidentiary standard it remains “in the noise.”
• Both hosts agree that existing copyright and antitrust doctrines cannot address every valid concern at issue here. Ben’s recurring answer is that policymakers must pass a new law, then openly debate its trade-offs, rather than stretch doctrines that “don’t really apply here.”
• An affirmative-permission regime might punish lawful AI developers while empowering actors willing to train on pirated or online material. Ben hedged that the UK had passed or was proposing something analogous—“I might be wrong”—but his realpolitik conclusion was categorical: only companies trying to comply would suffer.
• Heavy licensing obligations could entrench incumbents without reversing publishing’s existing structural decline. Ben said “the horse is halfway across the country,” while Andrew and Ben argued that burdens on AI startups would favor incumbents. Ben invoked YouTube, whose elaborate copyright machinery lets it pay rights holders and avoid abusive takedowns of videos with background music, but also means “there will never be another competitor to YouTube.”
• Ben still sees value in human media’s shared, one-to-many experience. Personalized AI is “all custom to you,” whereas tens of thousands hearing the same imperfect human podcast gain a common experience—an argument he admits is “totally talking my personal book.”

Deep dive

1. Andrew’s anti-monopoly instinct meets copyright’s limits

• Andrew remains instinctively uncomfortable with fair-use treatment: millions of uncompensated creators enable an industry dominated by companies worth roughly $5 trillion, transferring market power and future profits.
• He nevertheless leans toward fair use after reconsidering the copyright context: courts would be addressing speculative future harm, not demonstrated present harm, where the output itself is not infringing.
• The scale makes this unlike prior copyright disputes, but Andrew thinks the law cannot consistently prevent speculative market harm without a record of actual injury.

2. Chhabria reaches the right result—but Ben calls a better case illusory

• Andrew praised Chhabria for acknowledging threats to creators while ruling that these authors had not established lost market opportunities.
• Ben accepted the outcome but called the suggestion that plaintiffs could build a better case “a fantasy”: unmade purchases cannot be observed at scale, leaving genuine harm “in the noise.”

3. Valid policy concerns require new law

• Ben framed the dispute as a replay of their aggregator debates: antitrust could not fix those concerns, and copyright cannot fix these. A remedy needs new legislation.
• Andrew favored Chhabria’s warning against “robotically applying concepts from previous cases”; fair use should adapt to major technological change, despite Ben’s “living Constitution guy” tease.

4. Permission rules could reward evasion and incumbency

• Ben cautiously referenced a possible UK approach—“I might be wrong”—and considered a default requiring affirmative permission before copyrighted work enters training.
• His realpolitik objection: copyrighted works can remain available online or through piracy, so compliant developers bear the costs while unscrupulous competitors gain an advantage.
• Both argued that burdens on AI startups will not repair publishing’s structural decline and could favor incumbents. YouTube’s licensing system works, Ben noted, but its complexity means “there will never be another competitor to YouTube.”

5. Copyright protects incentives, while shared human broadcasting retains value

• Ben called copyright a “government-granted monopoly,” not a natural right: society accepts market distortion to encourage creation, but intellectual-property protections can also produce abuse, high prices, and counterproductive effects.
• Andrew contrasted creator safeguards with limiting Google, Meta, or Amazon take rates across hundreds of dependent industries.
• Ben’s point about media is that personalized AI is “all custom to you,” while a human broadcast retains value by giving many people one shared experience. He acknowledged that this is “totally talking my personal book.”