Suno's Hacked Source Code Confirms YouTube Stream-Ripping: Users Were Never Warned

A hacker who broke into Suno's internal systems last November handed investigators and journalists something the music industry's lawyers had been demanding in court for two years: the actual machinery of how the $5.4 billion AI music platform built the training corpus that now sits at the center of copyright lawsuits on three continents. The source code shared with investigative outlet 404 Media names specific platforms, records exact hour tallies, and shows that Suno used commercial proxy services — specifically, a company called Bright Data — to route scraping requests through rotating IP addresses and bypass YouTube's anti-bot protections . That detail matters beyond the copyright litigation. Under DMCA Section 1201, circumventing a technological measure that controls access to a copyrighted work is independently actionable — without proving any underlying infringement, and without a fair use defense.

The hacker, operating under the handle "ellie.191," also accessed a customer list comprising email addresses, phone numbers, and Stripe payment information for hundreds of thousands of Suno users. Suno learned of the breach in November 2025. It has not notified a single user.

The attack was not novel. Ellie.191 told 404 Media they gained access through the Shai-Hulud worm — a self-replicating npm supply-chain malware that Palo Alto Networks' Unit 42 had publicly identified as the first worm of its kind in npm history back in September 2025, two months before the breach. The worm works by compromising a developer's npm package installation pipeline, then harvesting GitHub tokens and cloud service credentials from the infected environment. Once the credentials are captured, the attacker replays them against the target's repositories and cloud infrastructure.

In Suno's case, ellie.191 used that path to reach private GitHub repositories and cloud services, pulling source code from 2023 and 2024 alongside the customer database. The motivation, the hacker told 404 Media: "I like to hack anything and everything."

Suno described the November incident as a "limited security incident that was quickly contained," said the exposed material consisted of "outdated source code that is no longer in use," and maintained that no sensitive personal information was compromised — citing the fact that it does not retain full credit card numbers. The company said the limited scope of the incident made formal user notification unnecessary.

Affected users who spoke with 404 Media had a different experience: they confirmed to the outlet that they had received no breach notification of any kind. Suno is headquartered in Cambridge, Massachusetts, where state law under M.G.L. c. 93H requires notification to any Massachusetts resident whose unencrypted personal information has been accessed by an unauthorized party. Email addresses and phone numbers qualify as personal information under that statute. The company's self-determined "limited breach" exemption is legally contestable.

The dataset inventories inside the leaked code make previously abstract claims concrete. A file labeled "youtube_music" recorded that the pipeline had ingested 2,013,545 music clips at the time of its last update. Separate dataset listed the following :

The total across these documented sources exceeds 380,000 hours — the equivalent of nearly 43 years of continuous audio. The code also shows Suno searching YouTube specifically for a cappella versions of songs, a deliberate methodology for capturing clean vocal tracks isolated from instrumentation. A separate pipeline used PodcastIndex to identify 420,000 podcasts with at least five episodes of 30 minutes each, and sought to download roughly one million hours of podcast audio.

The scraping instructions in one file directed the system to pull from "genius_hq, youtube_music, freesound, jamendo, imp, deezer, ytm_tagged," with a note that "non-music will be filtered out."

Suno has not disputed the code's contents. Its response pivoted to the "Original Creation, By Design" framing the company developed for litigation: the company intentionally excludes artist names from training metadata to discourage the model from replicating specific artists' styles, and it built detection filters to block user prompts that reference specific song titles or artist likenesses. "As we have stated in public filings and disclosures, Suno's AI models have been trained on publicly available music files and related metadata accessible on third-party websites on the open Internet," a spokesperson said .

Whether Suno's training constitutes fair use under 17 U.S.C. §107 is the central question before a federal court in Massachusetts — but it is no longer the only question the source code raises. The DMCA's Section 1201 anti-circumvention provisions prohibit bypassing any technological measure that effectively controls access to a copyrighted work, and they carry no fair use exemption. The circumvention itself is actionable, regardless of what happens to the content afterward.

YouTube operates a bot-detection and rate-limiting system specifically designed to prevent automated mass downloading of content. Using Bright Data's proxy network to route scraping requests through rotating IP addresses is precisely the kind of technical countermeasure the DMCA was written to prohibit — assuming a court agrees that YouTube's protections constitute a "technological protection measure" within the statute's meaning. The Recording Industry Association of America raised this circumvention theory in its September 2025 amended complaint against Suno; the hacked source code now offers the most direct evidence that proxy services were used specifically to defeat that system.

A federal court in the Udio case denied part of Udio's motion to dismiss the DMCA stream-ripping claims on May 21, 2026, meaning those claims survive and are proceeding to discovery. Suno has separately argued that its scraping "is not prohibited by the DMCA." The labels counter that "the violation lies in the circumvention, not the reason for it." That argument is now backed by code naming the proxy tool.

The practical significance: even if the Massachusetts federal court rules in 2027 that training generative AI models on copyrighted music is transformative fair use, that ruling cannot extinguish the DMCA anti-circumvention claims. Suno could win the copyright fight and still face separate liability for the means by which it acquired the training data.

The litigation picture has shifted since earlier press coverage suggested a July 2026 fair-use ruling in Massachusetts was imminent. The Massachusetts court's June 30, 2026 scheduling order moved dispositive motions to April 9, 2027, pushing the earliest possible US fair-use ruling back by more than a year. The actual ruling will come instead from Munich, where the GEMA v. Suno verdict is scheduled for July 31, 2026 , before the 42nd Civil Chamber of the Munich Regional Court — the same chamber and same judge who ruled for GEMA against OpenAI over song lyrics in November 2025.

In the Massachusetts case, Universal Music Group and Sony Music filed a motion on May 21, 2026 seeking to expand their complaint from the original 560 works to 61,026 specifically identified recordings — a figure their audio fingerprinting experts described as "only a small fraction" of what Audible Magic's content-recognition technology found in Suno's training data. At the statutory cap of $150,000 per work for willful infringement, those 61,026 recordings carry a theoretical maximum exposure exceeding $9 billion, up from the roughly $84 million implied by the original complaint. Suno opposed the expansion on June 4, citing a parallel New York court ruling that denied Sony the same expansion in the Udio case. Judge Saylor has not yet ruled on the motion.

Warner Music Group is no longer a party, having settled in November 2025 in an arrangement that included Suno acquiring Warner's Songkick live-music discovery platform and committing to launch licensed AI models with WMG's catalogue. Warner CEO Robert Kyncl framed the deal as an opportunity to "shape models that expand revenue." Universal Music and Sony are pressing forward.

Beyond the US, Denmark's collecting society Koda filed its own copyright action in Copenhagen City Court in November 2025 — Koda CEO Gorm Arildsen called Suno's practices "the biggest theft in music history." Koda's commissioned economic analysis projected that AI-generated music could reduce the Danish music industry's revenue by an estimated $680 million from 2025 to 2030, representing a 28% decline by decade's end.

All of this litigation was proceeding on the theory that Suno's training data included copyrighted recordings — a theory confirmed by Audible Magic's audio fingerprinting. The hacked source code goes further: it shows not just that the recordings were there, but how they were acquired.

The fair use doctrine's four-factor test has been the center of gravity in AI copyright litigation since the RIAA filed suit in June 2024. Suno's defense rests primarily on the First Circuit finding AI training transformative — a posture the June 2025 ruling in Bartz v. Anthropic in the Northern District of California partially supported for lawfully acquired training data, finding that kind of use "spectacularly transformative." The same ruling denied that protection to training on piracy-sourced material.

The proxy circumvention evidence the hack reveals may move Suno's training data acquisition from the "lawfully acquired" column toward something closer to the piracy-sourced column — not because the content itself was pirated, but because the means of acquisition potentially violated DMCA anti-circumvention law regardless of the content's legal status. That is the structural double exposure the leaked code creates, and it is what makes the training data methodology, and not just its scale, legally significant.

For the independent artists whose recordings are not covered by the major-label settlements, the path to damages depends almost entirely on how the court resolves these questions. A class action filed in the Northern District of California in late 2025, Nguyen v. Suno , alleges that roughly 60% of the 40 million tracks in Suno's training data came from independent artists who received no consideration from any settlement. If the Massachusetts court ultimately finds infringement liability, that ruling provides the legal foundation for those damages claims. If it finds fair use, it largely extinguishes them.

Yes, in a legally significant way. Most of the platforms Suno scraped — YouTube Music, Deezer, Genius — actively deploy bot-detection systems and rate-limiting to prevent automated mass downloading. The leaked source code shows Suno routed scraping requests through Bright Data's commercial proxy network to cycle through rotating IP addresses and evade those defenses. Under DMCA Section 1201, bypassing a technological measure that controls access to copyrighted works is independently prohibited — without a fair use exemption — regardless of what Suno ultimately did with the content. This is separate from and survives any resolution of the copyright infringement question.

The company says no. Suno describes the November 2025 incident as "limited" and maintains that not retaining full credit card numbers means no sensitive personal information was compromised. That legal conclusion is contested. Every US state has a data breach notification law, and Massachusetts — where Suno is headquartered — requires notification when a resident's email address or phone number is accessed by an unauthorized party. The affected customers who spoke with 404 Media confirmed they received no notification despite having their information and partial payment data accessed. Whether Massachusetts Attorney General enforcement follows is an open question.

Not as soon as earlier coverage suggested. The Massachusetts federal court case — UMG and Sony v. Suno before Judge F. Dennis Saylor IV — reset its schedule on June 30, 2026: dispositive motions are now due April 9, 2027. The earliest ruling in any major jurisdiction will come from Munich, where the GEMA v. Suno copyright verdict is scheduled for July 31, 2026. A GEMA win would be the first European ruling confirming that AI music training requires licensing authorization — and under German law, it could be enforced while Suno appeals.

Platform availability is the most immediate question. A GEMA victory in Munich can be enforced immediately under German law, potentially enabling an injunction against Suno's European operations. A US ruling against Suno on copyright infringement would not automatically shut down the platform, but it would create enormous financial pressure — the 61,026 recordings currently at issue carry a theoretical maximum of over $9 billion in statutory damages, and that figure represents only a fraction of what Audible Magic's fingerprinting found. Practically, a loss would likely force Suno to seek licensing deals similar to what Warner has already formalized, or face injunctions targeting the training data itself. The user experience on any post-ruling platform would reflect whatever licensing costs those deals impose.