Rights Holders Are Now Pricing AI Training One Song at a Time
Elvis Presley’s rights holders are suing Anthropic over 274 named songs, showing that a clean, countable catalog is now what turns an AI grievance into a priced claim.
Elvis Presley Rights Holders Sue Anthropic Over 274 Songs, Seeking About $41 Million
Authentic Brands Group, which owns Elvis Presley’s music and likeness, and other Presley rights holders sued Anthropic on Friday. They allege that 274 of his songs, including “Can’t Help Falling in Love” and “All Shook Up,” were used to train Claude without permission. They seek $150,000 per track, about $41 million.
The complaint reaches past training. It alleges Claude has produced infringing output of song structures and hooks even after guardrails were added, and that Anthropic knew the guardrails were ineffective and keeps infringing willfully. Another named example is “A Little Less Conversation.” The coverage calls it one of the first major suits brought directly by an artist’s estate against an AI developer. The suit follows claims against AI developers by the RIAA and by Sony, Universal and Warner. The coverage includes no response from Anthropic and does not name the court.
The number is the story: a rights holder that can name 274 titles and multiply by a damages figure has turned a grievance into a claim a court can price.
- A countable catalog.Per-track damages only work when works are identified, registered and matched to what was used. Rights holders with clean catalog data enter a negotiation with a number; those without it enter with a position. Catalog hygiene is now a revenue function, not an administrative one.
- Output as a second claim.The suit targets what the model produces as well as what it ingested, which turns a one-time training license into ongoing use that can be metered.
- Guardrails as evidence.Alleging that safeguards were known to be ineffective makes the safeguards themselves a subject of discovery, so what a developer records about its own filters becomes part of the price. I learned this building content attribution for broadcast radio: detection is never the hard part; markets form around a count both sides agree to trust.
Treat catalog data as the first licensing asset to fix, because the party that arrives with the count sets the opening price. Ask every licensee for the same per-work count in return, with dates attached.
- BuildDetection & TelemetryAudit your catalog for registration, identifier and ownership gaps, and start monitoring AI output for your most valuable titles so every future claim arrives with dated examples.
- LegalDeal ExecutionWrite per-work reporting, retention and deletion terms into any AI license under negotiation, so the count your contract relies on is produced by the licensee rather than assumed.
- CommercialEconomic ModelModel a per-work license rate well below the $150,000-per-track figure in this complaint, so a deal is the cheaper path for the developer and the count stays in your hands.
Suno Moves to Force the Mechanical Licensing Collective to Produce Streaming Data Dating to 2021
Suno has asked a court to compel the Mechanical Licensing Collective to comply with a document subpoena, and Udio has filed a near-identical request. The April demand sought monthly, label-by-label performance data from 2021 to the present for eleven streaming services, later narrowed to eight. The Collective says it does not collect label or distributor data directly, has no mandatory reporting fields for it, and that streaming services objected to disclosure. Suno says the Collective first agreed to produce several terabytes at a cost of about $5,000, then reversed. The Collective’s response is due October 30.
The Operator’s Take: AI developers are now asking for the same play-level data rights holders say they lack, which shows how much leverage sits with whoever holds the record of use.
Legal · Industry Positioning Review what usage data your distribution and collecting-society agreements actually give you, and add reporting rights at label, distributor and service level before a subpoena makes the gap public.
Spotify’s Direct Ad Sales Fall €53 Million as Automated Buying Lifts Ad Growth to 9%
Spotify’s ad-supported revenue reached €482 million in the third quarter, up 9% at constant currency and about 10% of total revenue of €4.97 billion. Automated channels, including auction-based biddable buying, added €91 million year on year while direct sales fell €53 million. Over nine months, automated channels gained €211 million and direct sales lost €177 million. The finance chief put biddable at up to 40% of ad revenue, but a company chart shows about a third. Spotify gave no numeric fourth-quarter ad guidance beyond growth approaching double digits.
The Operator’s Take: The growth is coming from the channel that is easiest to price and hardest to differentiate, so premium audio inventory needs measurement that survives the move to auctions.
Commercial · Economic Model Split your audio and video ad revenue by direct versus automated channel for the last four quarters, and set a floor price for auction inventory tied to the attention and completion data you can document.
DAZN Reports a $979.7 Million Net Loss as McDonald’s Ends Its 32-Year World Cup Sponsorship
DAZN’s first consolidated U.K. accounts, for the year to December 31, 2025, show revenue up 22.2% to $3.80 billion, an operating loss of $939.8 million and a net loss of $979.7 million. Sports rights cost $3.01 billion, and net liabilities nearly doubled to $2.41 billion. Owner Access Industries has provided more than $7 billion, though its support letter is not legally enforceable. Separately, McDonald’s is ending its 32-year FIFA World Cup sponsorship and shifting money toward intellectual property it owns, its chief financial officer saying the summer campaign underperformed expectations. The filing has no subscriber or viewing figures.
The Operator’s Take: Sports rights are inflating faster than the audience data that justifies them, and sponsors are redirecting money to properties they control.
Commercial · Deal Execution If you sell sponsorship around live rights, put owned or co-owned content inventory and outcome-based measurement into the renewal offer rather than defending a rights-fee premium.
Kochava’s StationOne Adds an Auto-Router and Spending Budgets to Rein In Enterprise AI Costs
Kochava added seven features to the enterprise tier of StationOne, its AI desktop hub. They include an Auto Router that picks the most cost-appropriate model for each request, token- or dollar-based budgets for users and teams with threshold alerts, usage analytics with projected costs, and reporting tied to business tasks. The release gave no prices, dates, savings figures or customer examples, and did not say whether budgets are hard limits. The coverage cites a KPMG survey of 2,145 leaders in which 49% had scaled back or paused AI agent deployments because costs outweighed value, and only 35% said AI costs were fully visible.
The Operator’s Take: Cost control is becoming a product category because few organizations can see what their agents spend. I learned this assessing AI readiness across media, finance and civic organizations: most AI programs stall on the operating model, not the model.
Build · Economic Model Set a per-team budget and a task-level cost report for every agent in production before scaling, and route routine requests to cheaper models first.