0% read

Your progress · 0/5 chapters

About 6 min total

CourtsISSUE #2 · STORY 19 OF 20Sep 18, 2026CONFIRMED

Music giants sue AI song maker Suno again

On Sep 18, 2026 Universal and Sony filed a second copyright suit against Suno over 60,202 recordings. Their claim carries a theoretical ceiling just above $9 billion, the largest asserted in an AI music case.

Illustration: a courtroom bench rendered as clean geometric oak volumes, with a grand gavel resting on its rail and a towering tree of branching musical notes rising behind it — every note a small glowing leaf,…

Read it your way

CHAPTER 1 · THE 60-SECOND VERSIONPicked for Explorers

A second, far bigger copyright suit

Two major labels are suing Suno again over a new AI music model. The first case covered 560 songs, while this one covers 60,202 recordings. At maximum rates the claim could top $9 billion.

Same judge, new caseBoth suits sit before Judge Saylor in Massachusetts.

Covers 60,202 recordingsThe claim is about 107 times larger than the first case.

A nine-billion ceilingAt $150,000 per willful work the theoretical cap passes $9 billion.

Nothing is provenThese are allegations, not court findings.

Finish this chapter for +15 XP

Deep dive

The full research, labeled and sourced

CONFIRMED24 sources · 68 min

Story identity

Evidence status: CONFIRMED (the filing itself is verified via the complaint PDF, docket record and multiple independent outlets); all factual allegations in the complaint are party claims (plaintiffs' allegations) and are labelled COMPANY CLAIM until adjudicated; Suno's response is labelled COMPANY CLAIM (company statement). No allegations have been adjudicated.

Core identity (FACT, CONFIRMED):

On Friday 2026-09-18, UMG Recordings, Inc. and Capitol Records, LLC (Universal) and Sony Music Entertainment and ten affiliated label entities filed a second copyright lawsuit against Suno, Inc. in the U.S. District Court for the District of Massachusetts, case 1:26-cv-14275, assigned to Judge F. Dennis Saylor IV — the same judge presiding over the labels' first case against Suno (1:24-cv-11611-FDS, filed 2024-06-24).
The complaint asserts 60,202 sound recordings — approximately 107x the 560 works in the first case — with a theoretical statutory-damages ceiling of just over $9 billion ($150,000 max per willfully infringed work, i.e. $9,030,300,000), plus up to $2,500 per act of circumvention (~$150,505,000 if applied once per recording), an injunction against further use/circumvention, and a jury demand.
The twelve plaintiff entities: UMG Recordings, Capitol Records, Sony Music Entertainment, Alamo Records, Arista Music, Arista Records, LaFace Records, Records Label, Sony Music Entertainment US Latin, Ultra Records, Volcano Entertainment III, and Zomba Recording LLC. Warner Music Group is not a plaintiff — it settled with Suno in November 2025 and is now a licensed partner.
The new complaint targets Suno v6 (launched 2026-09-09 with Warner, BMG and Believe as licensed partners), arguing that v6 was built on outputs and preference data from prior Suno models trained on the labels' recordings — "v6 is not a fresh start; it is the fruit of the same poisoned tree."
Suno answered publicly on 2026-09-22 (statement to media): the claims are "fundamentally flawed on both the facts and the law"; v6 was trained on "content licensed from our partners, interactions including creations and preference signals from our community, and the accumulated learnings from our team."

Relationship to the discovery record: the DISCOVERY_RAW.json S19 entry (case in Boston, Judge Saylor, 60,202 recordings, ~$9B ceiling, Suno responded Sep 22) is consistent with the verified record on every element. One refinement: the litigation tracker indicates the parties began coordinating the two cases' schedules (dispositive-motion dates pushed toward June 22, 2027), a detail confirmed after the discovery snapshot.

✓

What happened?

🎓 For Explorer

Chronology and context (FACT unless labelled):

2024-06-24 — The three majors (UMG, Sony, Warner via the RIAA) sued Suno in D. Mass. (1:24-cv-11611-FDS) over 560 "illustrative" recordings; Suno's defense: training on copyrighted music is "quintessential fair use." (CONFIRMED)
2025-11 — Warner Music Group settled and exited the case, becoming Suno's first major licensing partner (Suno also acquired Songkick). UMG and Sony remained as plaintiffs. (CONFIRMED)
2026-05-21 — UMG and Sony moved to add 61,026 additional works to the first case, identified via Audible Magic audio fingerprinting of Suno's training-data corpus (millions of match results). (CONFIRMED)
2026-08-12 — Suno signed BMG (its second major rightsholder deal). (CONFIRMED)
2026-08-18 — Judge Saylor deferred a ruling on the §1201(a) stream-ripping amendment but DENIED without prejudice the addition of the ~61,000 works: "simply adding claims involving 61,026 additional works to this lawsuit will have obvious consequences of complexity and delay," and the labels' fair-use entitlement was expressly preserved ("Plaintiffs are of course entitled to pursue valid copyright claims, and the magnitude of the alleged infringement is not a defense"); he pointed to a separate, parallel case, likely assigned to his own session. (CONFIRMED — court order)
2026-08-25 — Labels filed the second amended complaint in the first case adding the DMCA § 1201(a)(1) stream-ripping claim (Suno admitted in a 2026-09-01 filing that it used YT-DLP to obtain YouTube audio). (CONFIRMED)
2026-09-08 / 09-09 — Suno announced the Believe partnership and launched v6 (v6, v6 Wild, v6 Mini) "in partnership with WMG, BMG and Believe." (CONFIRMED)
2026-09-18 — UMG and Sony filed the second lawsuit (1:26-cv-14275) over 60,202 recordings, with three counts: direct infringement of post-1972 recordings; direct infringement of pre-1972 recordings (Music Modernization Act); and DMCA § 1201(a) circumvention (stream-ripping). (CONFIRMED — the filing itself)
2026-09-18 (same day) — Reporting: Bloomberg Law ("UMG, Sony Expand AI Music Fight Against Suno With 60,000 Songs," 9:29 PM UTC), TNW, MBW (obtained the full 45-page complaint), Variety, Hollywood Reporter, Music Week, Digital Music News, Engadget. The case appeared on PacerMonitor/Law360 trackers assigned to Judge Saylor. (CONFIRMED)
2026-09-22 — Suno's media statement responding to the new suit (see Section 1). (CONFIRMED — company statement) Reporting of the statement: MBW, Engadget, The AI Musicpreneur. (CONFIRMED)

What changed?

The music-AI copyright fight doubled in size overnight: 560 works in suit became ~60,762 across two parallel cases (560 + 60,202), with the second case alone carrying a >$9B theoretical damages ceiling — one of the largest statutory-damage exposures ever asserted in an AI copyright case.
The theory of liability moved "up the model chain": the labels are no longer only saying "you copied our songs into your training corpus." They now argue that successor models trained on the outputs and user-preference signals of infringing models inherit the infringement — a "poisoned tree" / laundering argument that targets the industry's standard practice of distilling and iterating on prior checkpoints.
Licensing presence flipped from defense to weapon: Suno's three rightsholder deals (Warner, BMG, Believe) are pleaded as proof "that a functioning market exists for licensing sound recordings to train generative AI models," directly pressuring Suno's fair-use defense on the market-effect factor.
A second front opened under the same judge: Saylor now runs both the fair-use-determined first case and the expanded second case, with coordination already under discussion (scheduling amendments pushing dispositive motions toward June 22, 2027 per the litigation tracker).
The DMCA § 1201(a) circumvention claim is now a standalone second-case count, not just an amendment in case one — doubling the injury theory (infringement + circumvention of YouTube's "rolling cipher" download protections via YT-DL/YT-DLP).
Consumer/user content is now live evidence: Suno's own confirmation (Sep 22) that v6 trained on user "creations and preference signals" puts user-generated tracks inside the litigation's factual core, alongside Suno's terms-of-service licence over user content.

↔

Before → Change → After

🎓 For Explorer

Before: One lawsuit (since June 2024) over 560 works; a fair-use question pending; Warner settled out; Saylor refused to let the labels expand the original case and pointed to a parallel filing; Suno signed Warner, then BMG, then Believe; Suno launched v6 (Sep 9) on licensed + user-derived data; the labels held an Audible Magic analysis showing "millions" of their works in Suno's training corpus.
Change (2026-09-18): UMG and Sony filed case 1:26-cv-14275 asserting 60,202 recordings and three counts (post-1972 infringement, pre-1972 infringement, DMCA circumvention), attacking v6's training lineage, Suno's licensing posture, its terms of service, its artist-name prompting, and its market impact — with a >$9B statutory ceiling and a jury demand.
After: Suno faces two parallel Saylor-run cases, a pending first-case answer/amendment fight over stream-ripping, a September 30 close of fact discovery in case one, a public acknowledgment that v6 ingested user creations, and a market reading the case as the new pricing floor for licensed music data. Adjacent fronts: Sony v. Udio II (30,117 recordings, filed July 20), GEMA's German win against Suno (July 2026), an independent-artists class action, a data-breach class action, Jamendo v. Suno, and a publishers' suit against Anthropic (Aug 31).

⚙

How it works

The complaint's legal machinery (all COMPANY CLAIM until adjudicated):

Count 1 — direct infringement (post-1972 recordings): Suno reproduced the 60,202 recordings into its training corpora without licence; statutory damages up to $150,000 per work if willful.
Count 2 — direct infringement (pre-1972 recordings): same theory for recordings protected under state-law-remnant rights preserved by the Music Modernization Act.
Count 3 — DMCA § 1201(a)(1) circumvention: Suno allegedly bypassed YouTube's "rolling cipher"/anti-download protections using stream-ripping tools (YT-DL/YT-DLP) to harvest audio — the same claim Saylor allowed into the first case on Aug 18; up to $2,500 per act.
The "poisoned tree" / laundering theory (v6): labels allege Suno trained v6 on (a) outputs of its prior models (v4, v4.5, v4.5+, v5, v4.5-all, v5.5, each allegedly trained on the unlicensed corpus) and (b) user preference signals about those outputs — plus, "on information and belief," knowledge distillation of a "teacher" model into a "student" model. Corporate quote: "Training a 'new' model on the outputs of an infringing model does not eliminate the infringement; it launders it." Because Suno has never said it destroyed its original copies, "v6 thus rests on a foundation of unauthorized copies."
Licensing-market argument: three deals in under a year (Warner Music, Nov 25 2025; BMG, Aug 12 2026; Believe, Sep 8 2026) prove a functioning market for training-data licences, undermining fair use's market-effect factor. The complaint quotes Suno CPO Jack Brody ("not in exchange for training", "not really about the data") and dismisses the characterization as "self-serving."
Knowledge/willfulness argument: Suno's own terms of service (since at least Jan 2024) require users to grant a worldwide, non-exclusive, fully paid-up, sublicensable, irrevocable licence over uploaded content — the labels argue this proves Suno knew training on someone else's audio requires permission, supporting the willfulness tier.
Market-harm evidence: Deezer (July 2026) — AI tracks passed half of daily uploads (~90,000/day, up from ~39% six months earlier); Suno's investor pitch deck claims users generate "an entire Spotify catalog's worth of music every two weeks"; The Velvet Sundown reached 1M+ monthly Spotify listeners before being exposed as AI.
Remedies sought: statutory damages to the ceiling, $2,500 per circumvention, an injunction against further infringement and circumvention, and a jury trial.

Why it matters

🎓 For Explorer

It consolidates the single most consequential open question in generative media: whether training (and iterating on) models over copyrighted sound recordings without a licence is infringement or fair use. A Saylor ruling in either direction — now accelerated by two parallel cases — sets the pricing floor for licensed music data across the entire AI-music industry.
The >$9B ceiling makes it one of the largest asserted copyright exposures in AI — an order of magnitude above the first case's ~$84M ceiling (560 × $150,000) and the $1.5B Anthropic book-settlement that reset expectations in 2025.
"Poisoned-tree" theory is a template risk for every model lab: if a court credits the argument that models trained on the outputs of infringing predecessors inherit liability, the industry's routine practice of iterative distillation becomes a liability chain, not a clean-slate defense. This is the week's clearest legal signal on AI training lineage.
It frames the licensing market explicitly: the labels turned Suno's own Warner/BMG/Believe deals into evidence that a functioning market exists — i.e., that AI companies can and should pay. That framing travels directly into every AI-copyright negotiation in music, books, news and video.
It joins the week's accountability wave: in the same window California signed a frontier-AI kill-switch EO (S04), BC sued OpenAI (S06), the UN briefed on losing control of agents (S05), and Amazon amended its Perplexity suit (S15) — S19 is the copyright-music vector of the same enforcement turn.

✦

What became possible?

🎓 For Explorer

A consolidated legal answer on music-AI training sooner: one judge, two cases — a fair-use-first case and a broad works case — can produce complementary rulings (fair use on the merits in case one; statutory-damages/scope questions in case two) without the delay Saylor cited when he denied the amendment.
Discovery into Suno's model lineage: case two invites discovery into v6's training data, preference signals, distillation practices, and the retained copies of the original corpus — the "poisoned tree" facts.
A market-price discovery mechanism: the case's licensing-market argument creates pressure to surface what Warner/BMG/Believe actually paid — the exact discovery Suno fought to keep from UMG/Sony in the first case (Levenson's April 2026 ruling blocked the Warner settlement terms).
Precedent for suing the successor model: any plaintiff whose works were in model vN can now plead that vN+1 "laundered" the infringement — directly applicable to Udio, and by analogy to text, image and video generators.
A damages-scale benchmark: the ~$9B ceiling gives rights-holders a headline number for negotiating (or litigating) licensing rates equivalent to the $1.5B Anthropic benchmark for text.

◎

Implications

Technical

Audio fingerprinting as litigation evidence: the case is built on Audible Magic identifications inside Suno's training corpus (millions of match results; 60,202 asserted with registration verification). Content-ID-scale auditability of training corpora is now a first-class legal requirement, not a nice-to-have.
Training lineage disclosure: the "poisoned tree" argument makes model genealogy (which teacher models, which synthetic outputs, which preference signals fed the student) a legally material artifact. Labs without reproducible lineage records face an evidential handicap.
Preference-signal data is now legal evidence: Suno's confirmation that v6 trained on user preference signals ("which of two generations did the user pick") makes even non-audio interaction telemetry part of the infringement calculus.
Circumvention forensics: the § 1201(a) claim turns YT-DL/YT-DLP usage, rolling-cipher bypass mechanics, and download logs into technical facts in dispute — Saylor already noted the parties disagree about how those tools operate.
For audio modelers generally: the "fair use" versus "infringing lineage" distinction suggests the industry will bifurcate into licensed-data models and unlicensed-data models — with provenance tooling (dataset cards, fingerprint screening, licensing registries) as the emerging compliance layer.

Developer

If you build with Suno (or any unlicensed-data music model): your product's legal posture is now contested at a >$9B scale; watch for injunctive relief that could constrain service availability or training-data retention.
If you build your own audio model: treat dataset provenance as a legal artifact — record fingerprint checks (Audible Magic-style), licence documentation, and model-lineage (teacher/student, synthetic-data, preference-signal usage) in machine-readable form; the complaint reads like a checklist of what plaintiffs will ask for.
If you fine-tune or distill any model that touched copyrighted media: the poisoning theory is not limited to music; document whether your base model was trained on unlicensed data and what your "fresh start" claim rests on.
If you use Suno's API for products: Suno's own terms grant it a licence over content you generate ("creations"); the litigation confirms those creations are being used as training signals — a material disclosure for any commercial product built on generated audio.
Do not read the complaint as settled law: nothing is adjudicated; the poisoned-tree theory faces serious hurdles (causation, independent creation, fair use), and Suno's fair-use defense remains live in case one.

Enterprise

Media and entertainment companies: the case defines the negotiated price of training data. If labels prevail on the licensing-market argument, expect catalogue-licensing rates for AI training to harden across music, publishing, film and news.
Enterprises building internal audio/music AI: require vendors to warrant licensed training data and model lineage; contract for indemnification against exactly these claims (60,202 works, $9B ceiling — the numbers mark the risk watermark).
Platforms hosting UGC from AI tools: the "user creations as training data" fact pattern raises disclosure obligations: users whose tracks get ingested into a model should know their work is training material — and may be entitled to a share if the model monetizes.
Streaming services and distributors: the Deezer half-of-uploads statistic and the market-harm argument directly implicate streaming-fraud economics; expect label pressure on platforms to filter/flag AI audio.
Legal/insurance: AI-training-dataset litigation is now a priced risk class; cyber/errors-omissions policies and diligence checklists will start asking for fingerprint-verified corpora and lineage documentation.

Strategic

For the labels: the two-case strategy is a deliberate legal architecture — keep the fast fair-use question in case one, open the big damages and circumvention front in case two under the same judge — a template (mirroring Concord v. Anthropic I/II) for managing AI copyright without waiting for legislation.
For Suno: the company chose the licensing haven (Warner/BMG/Believe) and still got sued — proving that partnerships do not shield lineage. Its valuation ($5.4B post-money on a >$400M raise) now carries a formal >$9B claim overhang; its "v6 is trained entirely from scratch" claims collide with its own Sep 22 statement naming user creations as training inputs.
For the AI-music sector: the case splits the industry into "licensed-data" and "scraped-data" camps; licensing deals become both revenue and liability-shield — yet the lawsuit shows even a licensed shield may not cover predecessor models.
For the broader model economy: "train once, iterate forever" is now legally risky; the poisoned-tree theory, if credited, applies to text and image models too, sharpening the divide between open-weight models with disclosed data and closed models with undisclosed corpora.
For investors: AI-music investments now price litigation risk explicitly; the German GEMA loss, the Udio II case, and this Suno II case are converging on a global answer that no amount of Western regulatory delay can postpone.

⚠

Risks & limitations

Risks

For Suno: an adverse ruling on the poisoned-tree theory (or a finding of willfulness) could produce damages beyond the company's ability to pay at the statutory ceiling; injunctive relief could force retraining of v6; the "creations" admission supplies the labels with a concededly user-derived training corpus whose outputs trace back to prior models. Fact discovery in case one closes Sep 30 — near-term pressure.
For the labels: a ruling that training is fair use (or that § 1201(a) does not reach YT-DL/YT-DLP as Saylor himself flagged uncertainty) would normalize the costless use of audio corpora; a "poisoned tree" theory stretched too far could be rejected and weakened by overreach; pre-1972 claims and willfulness tiers face evidentiary hurdles; the parallel-case strategy could still be consolidated or stayed.
For Warner/BMG/Believe: their deals with Suno are now litigation exhibits; if Suno loses, partner catalogs' value as "licensed data" may be questioned; if the v6 lineage is found tainted, the partners bought into a poisoned tree.
For users/artists: user "creations" are assertedly inside v6's training data; artists whose Suno-made tracks trained the model have unknowable exposure and no compensation channel — while being cited by both sides (market harm vs. community input).
Systemic: a multi-billion-dollar statutory award to legacy labels could chill AI-music innovation and investment; conversely, an industry-wide fair-use win would devalue catalogs for AI training and cut the labels' revenue thesis. Either way, the outcome sets a global benchmark.

Limitations

Unadjudicated allegations: every infringement, circumvention, willfulness, and poisoning claim in the complaint is a plaintiff assertion (COMPANY CLAIM status), not a finding.
PACER paywall: the full docket (all filings, the answer in case two, the scheduling order) was not directly accessed; case metadata (filed 2026-09-18, cause 17:101, judge Saylor, 14 parties) verified via PacerMonitor and Law360 public trackers; the complaint text verified via the MBW-hosted PDF (45 pages, Dkt. 1) and detailed press reporting.
Work-count variance: the first case's denied motion referenced 61,026 additional works (per Saylor's order and MBW); the proposed amended exhibit text references "approximately 60,304"; the filed case-two complaint asserts 60,202. The figures differ across filings — likely due to list refinement between May and September — and this research records 60,202 as the operative pleading number (consistent across TNW, MBW, Music Week, Bloomberg Law, DMN and the tracker).
Suno's Sep 22 statement is a company statement to media (per MBW/Engadget), not a court filing; Suno had not filed a formal answer to the new complaint as of the last consulted sources.
WSJ and Billboard coverage (cited generally in the discovery record) were not directly accessed; Bloomberg Law, TNW, MBW, Variety, THR, Music Week, DMN and Engadget provided verification instead.
No forensic verification of the "millions of recordings" claim: the Audible Magic analysis is described in the parties' filings; this research did not independently inspect the training corpus.

Open questions

Will the two cases be coordinated (scheduling, discovery, possibly consolidation), and does the June 2027 dispositive-motion date hold?
What exactly does Suno's formal answer in case two say about v6's training lineage, and does it walk back or clarify the Sep 22 "creations" statement?
Does the Court credit the poisoned-tree/laundering theory, and what evidentiary standard applies to finding v6 "derived from" infringing models?
What did Warner/BMG/Believe actually pay, and does the licensing-market argument survive the earlier ruling shielding the Warner settlement terms (Rule 408)?
Does § 1201(a) reach YT-DL/YT-DLP usage where the tools may access content through authorized means (Saylor's own caveat)?
How does the Sep 30 close of fact discovery in case one interact with case two's discovery?
What is the actual overlap between the 560 works, the denied 61,026, and the asserted 60,202 — and why do the counts differ?
Do streaming platforms (Deezer, Spotify) respond to the market-harm evidence with mandatory AI-filtering, and does that change the damages narrative?

→

What should you do with this?

Circle 1

Circle 1 impact and recommended action

Circle 1 (directly involved): UMG Recordings, Capitol Records, Sony Music Entertainment and the ten label plaintiffs; Suno, Inc.; Judge F. Dennis Saylor IV; the parties' counsel (Cloherty & Steinberg appear on the docket); Warner Music Group, BMG and Believe (licensees whose deals are now exhibits); Suno's investors; Suno users whose "creations" are asserted to be training data.

Recommended action (for the direct actors to consider):

Suno: file the answer promptly; reconcile the "trained entirely from scratch" launch narrative with the Sep 22 statement; prepare model-lineage and retention-policy discovery responses (case-one fact discovery closes Sep 30); consider a Daubert-grade position on fingerprinting and distillation evidence.
Labels: lock in the case-two scheduling order; coordinate discovery with case one; preserve the willfulness theory via the terms-of-service and artist-name-prompting exhibits.
Warner/BMG/Believe: prepare positions on their deals being characterized as "a functioning market" — that framing is favorable to their pricing but adverse to Suno's defense.
Users/artists: audit whether Suno-generated "creations" are being used as training signals and what the current terms actually license.

Circle 2

Circle 2 impact and recommended action

Circle 2 (adjacent): Udio/Uncharted Labs (Sony v. Udio II pending), GEMA and European collecting societies, other music-tech and AI-music startups, streaming platforms (Deezer, Spotify), publishers' AI litigation (Anthropic/Amodei case), law firms and e-discovery providers, insurers and reinsurers, music-industry unions and artist-rights organizations.

Recommended action:

AI-music startups and API developers: implement licensed-data warranties and lineage documentation before contracting — not after.
Streamers: prepare for label requests to filter AI audio; the Deezer 50%-of-uploads statistic is now a court exhibit.
Insurance: begin underwriting AI-training-data risk with fingerprint-audit conditions; the $9B ceiling is the new upper bound to model.

Circle 3

Circle 3 impact and recommended action

Circle 3 (broader ecosystem): all generative-AI labs and model developers (text/image/video/audio), open-weight providers and their disclosure practices, enterprises using AI-generated media, regulators (California EO S04, US AI Force S02, EU AI Act, Canadian legislation), capital markets, and the global licensing ecosystem.

Recommended action:

Labs: publish dataset provenance and model genealogy as a matter of course; a court here will define the standard of care for "successor model" liability.
Regulators: use the case's licensing-market evidence to design market-based AI-copyright solutions (mandatory licensing regimes) before courts set the price case-by-case.
Enterprises: add "training-data provenance + lineage warranty" to every AI-vendor contract, mirroring the S16 evaluation-deal discipline.

Business value

Business-value opportunities where genuine

Training-data provenance and lineage tooling: fingerprint-audit (Audible Magic-style), dataset cards, model-genealogy registries, and preference-signal logging — marketable to every audio (and general) model developer; the complaint is effectively a spec sheet.
Licensing-negotiation advisory: the case establishes a pricing floor for catalogue licensing; rights-holders and AI companies both need valuation and deal-structure counsel.
E-discovery and litigation analytics for AI cases: organizing training-corpus analyses, fingerprint results, and distillation records is a specialized, growing practice (S06's Tumbler Ridge docket is the same category).
AI-music compliance as a service: for platforms and API users — disclosure of training-signal usage, terms-of-service alignment, and audit-ready documentation.
Insurance products: AI-training-data indemnity and errors-omissions coverage with audit conditions is an open market.
For artists/users: tools and services that detect and monetize user-generated training data (a version of the "right to know and be paid for your creations being training signals").

Hands-On

Hands-on recommendation

Lab: COMPARE — see labs/S19.md. Practical exercise: compare the two Suno cases (and Suno I vs Suno II complaints) side by side — parties, works asserted, counts, damages theories and the procedural path from Saylor's Aug 18 order to the Sep 18 filing — and independently recompute the statutory-damages math (verified here: 60,202 × $150,000 = $9,030,300,000; 60,202 × $2,500 = $150,505,000; combined ~$9.18B vs. the first case's 560 × $150,000 = $84,000,000). It is a genuinely useful exercise because the case's entire strategic shape is a comparison the labels designed: keep the small case fast, make the big case wide.

↗

What happens next?

🎓 For Explorer

Weeks (late Sep–Oct 2026): Suno's formal answer in case two; case-one fact discovery closes Sep 30; scheduling-order amendment (parties asked to align both cases, dispositive motions reportedly pushed to ~June 22, 2027); continued Suno v6 user-backlash coverage and possible further artist suits.
Months (Q4 2026–2027): motion practice on the poisoned-tree theory and § 1201(a); discovery into v6 lineage, preference signals, and retained copies; possible consolidation or coordinated discovery orders from Saylor; Sony v. Udio II and GEMA developments in parallel; possible settlements involving licensing components.
Outlook (PREDICTION, medium confidence): the first-case fair-use ruling likely lands before the second case's merits, and Saylor has already signaled he wants a timely fair-use answer; a fast fair-use decision would frame case two's damages phase. The likeliest industry inflection is not a $9B judgment but (a) a fair-use ruling that defines the license/pirate boundary, or (b) a licensing-rich settlement that effectively prices the entire music-AI market. Either path raises the cost of unlicensed music training industry-wide and accelerates adoption of licensed-data pipelines.

★

Editorial takeaway

🎓 For Explorer

The first Suno case was about 560 songs and a principle. The second is about 60,202 songs and a business model — and a theory with a name that will haunt every model lab: "the fruit of the same poisoned tree." In one 45-page complaint, the labels turned Suno's own strategy against it: its licensing deals prove a market exists, its terms of service prove it knew permission was required, its users' "creations" prove v6 inherited what v1–v5 learned, and its Spotify-catalog-every-two-weeks output proves the market harm is "present, measured, and accelerating." Nothing is adjudicated — every claim is still an allegation — but the ledger is now public: a ~$9.18B theoretical ceiling, one judge, two cases, and a whole industry watching whether the next model Suno trains will have to be born genuinely clean.

Evidence labels used: CONFIRMED (independently corroborated events and documents), COMPANY CLAIM (party/company assertions, unadjudicated), INDEPENDENT EVIDENCE (media/court-derived verification), EARLY RESEARCH (reported but contested/undetermined), INTERPRETATION/PREDICTION (analyst views). The complaint's factual allegations are allegations, not findings; Suno's and the labels' statements are attributed and labelled.

Illustration: frame: a courtroom bench of clean geometric oak with a grand gavel on its rail; behind it rises a tree whose branches carry glowing note-shaped leaves — an artistic impression of a second copyright…

⌘

Lab: COMPARE

Step 1 — Build the Suno I vs Suno II comparison table

Using the sources in sources/S19.md (complaint PDF, Saylor's Aug 18 order, Justia/PacerMonitor dockets, MBW/TNW reporting), fill in:

Dimension	Suno I (1:24-cv-11611-FDS)	Suno II (1:26-cv-14275)
Filed	2024-06-24	2026-09-18
Judge	F. Dennis Saylor IV	F. Dennis Saylor IV
Plaintiffs	UMG/Capitol/Sony + Warner labels (Warner later settled/exited)	UMG/Capitol/Sony + 10 label affiliates (12 entities; no Warner)
Works asserted	560 ("illustrative sampling")	60,202 (Audible Magic-identified)
Counts	Direct infringement	3 counts: post-1972 direct; pre-1972 (MMA); DMCA § 1201(a)(1) circumvention
Model generation attacked	v1-era training corpus	v6 lineage via prior models (v4, v4.5, v4.5+, v5, v4.5-all, v5.5) + distillation
Damages ceiling (recomputed in Step 2)	$84,000,000	~$9,180,805,000 combined
Key procedural origin	—	Saylor's 2026-08-18 denial without prejudice + invitation to file separately

Verification targets: (a) same judge on both; (b) Warner absent from Suno II; (c) 60,202 vs 560 (≈107.5x); (d) the three-count structure; (e) the poisoned-tree/laundering theory as the new Suno-II element.

Step 2 — Recompute the statutory-damages math (verified here)

Standard ceiling $150,000/work (willful) and $2,500/act (circumvention). Run:

60,202 × 150,000 = 9,030,300,000  (≈ $9.03B — "just over $9 billion" in press)
60,202 × 2,500   = 150,505,000   (≈ $150.5M if applied once per recording)
combined          = 9,180,805,000 (≈ $9.18B theoretical ceiling)
560   × 150,000  = 84,000,000    (Suno I ceiling — matches MBW's ~$84M figure)

Expected result: the ">$9B ceiling" claim checks out exactly; note it is a theoretical statutory maximum, contingent on willfulness findings and per-work/per-act application — no court has found liability (Section 13 of research/S19.md).

Step 3 — VERIFY the "poisoned tree" chain against the complaint PDF and Suno's Sep 22 statement

From the complaint: extract the three-step allegation (prior models trained on unlicensed corpus → users generate tracks and preference signals with those models → v6 trained on those creations/signals and, on information and belief, distilled teacher→student). Quote-check: "v6 is not a fresh start; it is the fruit of the same poisoned tree."
From Suno's Sep 22 statement (via MBW/Engadget): extract the four claimed v6 training sources ("licensed partner content", "creations and preference signals", "team learnings").
Cross-check the tension: launch-day claim ("trained entirely from scratch", "doesn't include data from Universal or Sony" — Brody) versus the Sep 22 acknowledgment of user "creations" in training. Expected outcome: the two statements are in tension on the "creations" item — this is exactly the fact pattern the labels plead.

Step 4 — Trace the procedural chain

From Saylor's Aug 18 order (Dkt. 272, CourtListener PDF): 560 original works → motion to add 61,026 (denied without prejudice) → invitation to file separately → Suno II filed Sep 18 with 60,202 works, assigned to Saylor. Note and record the count variance (61,026 requested vs ~60,304 in the proposed exhibit text vs 60,202 asserted) as an open question.

Expected outcome

A completed comparison table (Suno I vs Suno II), a verified damages computation log (≈$9.03B infringement ceiling, ≈$150.5M circumvention add-on, ≈$9.18B combined), a two-column quote extract of the poisoned-tree chain vs Suno's Sep 22 statement, and the procedural timeline. This leaves a reusable "case-vs-case" audit pattern for the parallel legal stories in the same window (S15 Amazon v. Perplexity, S06 BC v. OpenAI) and demonstrates why the damages headline must always be read as an unadjudicated ceiling.

Time estimate

~45–60 minutes.

≡

Research sources

Primary Sources (8)

Primary

Suno — Terms of Servicethe complaint's willfulness argument — Suno's user-content licence (worldwide, non-exclusive, fully paid-up, sublicensable, irrevocable, covering reproduction/modification/derivative works); version in effect since 2026-09-03 covers inputs and outputs including model improvement (per The AI Musicpreneur's reading of the same document). — primary document (CONFIRMED existence and general content; precise clause text per complaint and AI Musicpreneur quotes).Date: version cited in effect 2026-09-03; licence language present since at least January 2024 (per complaint and coverage)

Music giants sue AI song maker Suno again

Read it your way

A second, far bigger copyright suit

From 560 songs to 60,202

Follow what Suno files

Things to keep an eye on

Did it stick?

Does a model trained on an older model's output carry the same legal risk?

The full research, labeled and sourced

What happened?

What changed?

Before → Change → After

How it works

Why it matters

What became possible?

Implications

Technical

Developer

Enterprise

Strategic

Risks & limitations

Open questions

What should you do with this?

What happens next?

Editorial takeaway

Lab: COMPARE

Expected outcome

Time estimate

Research sources