Sony accuses Anthropic of 'brazen campaign' to train Claude on its music — and wants up to $150,000 a song

Sony Music Publishing and Warner Chappell Music sued Anthropic in Northern California, alleging it illegally scraped/torrented thousands of copyrighted songs to train and operate the Claude AI series. The complaint cites a June 2025 ruling that Anthropic downloaded over 7 million pirated books for Claude and claims Claude can output near-identical copyrighted lyrics, with relief sought including statutory damages up to $150,000 per composition used for training. The escalating copyright litigation adds legal and potential financial overhang for Anthropic’s AI rollout.

Analysis

This is less about one lawsuit and more about the market repricing the cost of training data. If courts keep validating large statutory-damages claims, frontier model vendors face a creeping tax on gross margin and product cadence: more licensing, slower dataset expansion, and a higher hurdle to justify today’s revenue multiples. The second-order winner set is broader than the named plaintiffs—any proprietary-content owner with searchable archives (news, music, books, image libraries) gains bargaining power as the “free data” era becomes less defensible.

For NYT, the direct P&L impact is likely immaterial in the near term, but the optionality matters. Every successful rights-holder case improves the expected value of future settlements and makes archive/content licensing more credible as a revenue line, even if it is small versus subscriptions. The cleaner read-through is valuation support for premium content owners rather than a fundamental earnings upgrade; the market may overread this as an immediate monetization event when it is really a long-dated bargaining shift.

The main risk to the thesis is judicial narrowing: if courts separate training from output liability or cap damages at a level that is easily absorbed, this becomes headline noise and the AI multiple overhang fades. Near term, the trade is sentiment-driven over days to weeks; the real catalyst path is 1-3 months of discovery motions and additional filings that either widen or contain the legal blast radius. Over 6-18 months, the structural outcome is a licensed-data regime that benefits rights holders and penalizes pure-scrape AI economics.

Key Decisions for Investors

• No standalone NYT trade on this headline alone; the implied equity impact is too indirect to justify a re-rating position before earnings or a direct NYT AI monetization disclosure.
• If AI copyright headlines continue to stack up, consider a small relative-value long NYT / short QQQ pair on weakness, targeting legal-optionality outperformance over 1-3 months; invalidate if courts narrow damages or a broad licensing framework reduces uncertainty.
• Watchlist trigger for NYT: any explicit guidance on AI licensing, archive monetization, or litigation reserves on the next 1-2 earnings calls; only then would this become a tradable fundamental catalyst.
• For broader exposure, avoid chasing AI beta until there is clarity on data-cost inflation; the better short is any name whose valuation assumes frictionless model scaling and low training-cost permanence.