News & Media

Reservoir Sues Two AI Developers While Meta Shifts AI Ad Risk to Advertisers

Written by Paul E. Wright | Oct 23, 2026, 4:00:00 AM

Reservoir’s suits against Suno and Anthropic turn signed AI licenses into evidence and put model weights in the remedy, while Meta moves AI ad risk onto advertisers.

Regulatory, Litigation & Legislative Track

Reservoir sues Suno over 1,050 songs and Anthropic over 853 lyrics

Independent music rights company Reservoir filed two copyright suits on October 21 in the U.S. District Court for the Northern District of California. The Suno complaint covers 1,050 compositions and seeks statutory damages of up to $150,000 per willfully infringed work, $157.5 million, plus DMCA counts, for just over $186 million in total. The Anthropic complaint covers the lyrics of 853 compositions and seeks just over $151.4 million.

Reservoir alleges Suno used proxy services to scrape songs and lyrics from YouTube Music, Deezer and Genius, citing source code reported on in July, and trained on cover versions alongside originals. It cites Suno’s licensing deals with Warner, BMG and Believe as evidence of a functioning market. Both suits also ask for injunctions and the impoundment or destruction of copies, including those in training datasets and model weights.

Read the full article →

The Operator’s Take

The complaint turns the licensing deals Suno has already signed into the plaintiff’s best evidence, and it names model weights as part of the remedy.

  • Licenses as proof of market.Pointing to deals with other rights holders is aimed at the fair-use question of whether a licensing market exists, so every signed deal now doubles as evidence for the next plaintiff.
  • Weights in the remedy.Asking for destruction of copies inside training datasets and model weights moves the dispute from damages to what the model may keep, which changes a developer’s calculation at settlement.
  • Arithmetic that scales.At up to $150,000 per willfully infringed work, 1,050 songs becomes a nine-figure number before any DMCA counts, so catalog size, not individual song value, drives exposure.

I learned this running embedded finance programs across a dozen countries: partners accept a split only when every one of them reads the same record. Rights holders who can produce a dated record of what was used, and AI companies that can produce the matching one, will settle and license faster than those arguing from separate stories, and the side without its own record will be negotiating from the other side’s.

  • LegalDeal Execution
    Use every signed license in your catalog as evidence of market value, and make sure each one records scope and rates in a form you can produce in discovery.
  • BuildDetection & Telemetry
    Document which of your works appear in scraped sources and which platforms were the route, so claims rest on your own records rather than a plaintiff’s source-code leak.
  • CommercialEconomic Model
    Price training, output and weight-retention rights as separate lines, since the remedy sought now treats them separately.
Regulatory, Litigation & Legislative Track · 22 October 2026

Suno admits it has not deleted all the recordings used to train its AI, in answer to UMG and Sony’s laundering lawsuit

In a 44-page answer filed October 20 in the U.S. District Court for the District of Massachusetts, Suno acknowledged it has not deleted all recordings used to train its models, saying retention is consistent with the labels’ earlier request that it keep the training data sets. The labels’ September 18 suit alleges its v6 models were trained on outputs of earlier, allegedly infringing models, which they call laundering. Suno denies each model used the same corpus, denies the artist-name ban was a “fig leaf,” and argues any copying is fair use. It answered rather than moving to dismiss.

The Operator’s Take: A preservation duty and a deletion duty now point in opposite directions, and licenses that ignore the conflict will not survive discovery.

Legal · Deal Execution Write retention, deletion and certification terms into every AI license, with a named party who certifies and audit rights for you.

Read more →

AI Content Licensing, Provenance & Deal Intelligence · 22 October 2026

UMG becomes first major music company to license TYGER, which uses AI to adapt songs to gameplay in real time

Universal Music Group has become the first major music company to license TYGER, a San Diego service that lets gamers run music alongside games with licensed tracks adapting to live gameplay using patented technology. Select recordings and songs from UMG’s recorded and publishing catalogs will be on the paid tier, and some recordings on the free, ad-supported tier. TYGER’s library holds more than 400,000 songs, mainly independent, and the beta runs across more than 20 games on PC. Early access to the paid tier costs $3.99 a month, rising to $5.99. Financial terms and consent mechanics were not disclosed.

The Operator’s Take: Adaptive music is a new use type, and the first major license sets the template before anyone has published a rate.

Commercial · Economic Model Define real-time adaptation as a separately priced use in your catalog terms before the next game-audio request arrives.

Read more →

Product & Platform Strategy · 22 October 2026

Meta’s AI rewrites the ads. From October 30, advertisers carry the risk

Meta’s revised Commercial Terms add an AI Features section effective October 30. Adweek reads it as making advertisers solely responsible for changes Meta’s AI makes to their ads, including features Meta turns on itself; the publication reporting this could not verify that against Meta’s preview. Earlier wording applied advertiser liability only when the advertiser chose Meta’s AI creative tools. Continued use counts as acceptance, Meta’s own liability is already capped at the greater of $100 or fees paid in the prior 12 months, and one bookseller found an AI-generated illustration it never requested in its ad.

The Operator’s Take: The same notice-and-continued-use mechanism that carries the crawling clause is moving AI risk onto the counterparty.

Legal · Deal Execution Review ad agreements for AI-edit indemnity before October 30, switch off any default AI enhancements the platform lets you, and require human approval for anything Meta alters.

Read more →

Harness, Workflow & Agentic Execution · 22 October 2026

Q3: Omnicom’s media business is growing, but is it agent-proof?

Omnicom reported Q3 core revenue of about $6 billion, up 7.2% organically. Integrated media made up 53% of core revenue and grew more than 10% organically, and the U.S., about 60% of core revenue, grew 10%. Executives credited the World Cup and, to a lesser degree, the midterm elections, and warned neither repeats in 2027. It is building an AI marketing intelligence platform called Omni. AdExchanger frames the open question: much of the media buying and content work the group earns from is work AI agents are being built to do.

The Operator’s Take: Growth driven by one-time events and a platform bet is not proof that the intermediary layer survives agent-led buying.

Commercial · Economic Model Model how much of your agency-intermediated revenue is exposed to agent buying and move fees toward outcomes you can measure.

Read more →

AI Content Licensing, Provenance & Deal Intelligence · 20 October 2026

WGA East partners with Cloak to give writers an AI shield

The Writers Guild of America East is partnering with Cloak, an AI-protection software company, on a new member offering. Cloak’s patent-pending software makes documents unreadable to most AI models while keeping their formatting and readability for human viewers, and a registrant can send an AI-proof copy of a file to anyone they choose. Guild members get free access, and non-members can add Cloak protection when they register a work through the guild’s script registration service for an extra $10 per registration. The guild says members need not worry that work sent to producers or agents will train AI models.

The Operator’s Take: A union now treats technical protection as a member benefit, but a vendor’s claim that documents are unreadable to AI is a hypothesis until someone tests it.

Build · Detection & Telemetry Test any AI-proofing tool against the crawlers and models you care about, log the results, and treat the vendor’s claim as unproven until you have.

Read more →

Harness, Workflow & Agentic Execution · 22 October 2026

Spotify Q3: revenue up 16% as the AI remix tool waits on rights partners

Spotify reported Q3 revenue of $5.57 billion, up 16% year on year, with 306 million paid subscribers and 788 million monthly active users. Operating income was $753 million, and the stock rose about 9% after the report. Executives were noncommittal on release timing for the AI remix and cover tool, though it has added Believe and TuneCore as partners for remixes and covers. A co-chief executive said the goal is a version made “with consent, credit, and compensation at the core.” Executives also called an agentic media platform a long-term opportunity.

The Operator’s Take: The remix tool is gated by rights partners rather than technology, which is where consent, credit and compensation get priced.

Commercial · Deal Execution Ask for explicit remix and cover consent terms, with credit and payment mechanics, in distributor and label agreements now.

Read more →