Suno Leak: AI Music Trained on Millions of Scraped Songs

What Happened?

A hacker accessed the internal repositories of Suno, one of the most popular generative AI music startups, and leaked fragments of its source code. According to The Next Web's analysis, the files reveal that Suno trained its model on millions of songs and lyrics scraped from the web without explicit permission from rights holders. The code includes references to massive databases of songs scraped from platforms like Spotify, YouTube, and lyrics sites. The leak, which occurred in July 2025, exposes a dataset estimated to exceed 50 million songs, including works by artists like Taylor Swift, The Beatles, and Drake, whose record labels have been particularly active in defending copyright.

Why It Matters

This case is not isolated. In 2023, Getty Images sued Stability AI for using its photos without a license, and in 2024, The New York Times sued OpenAI and Microsoft for copyright infringement. Now, the Suno leak confirms the suspicions of many artists: generative AI has been feeding on protected works without compensation. The difference is that Suno operated in a legally gray area, and this evidence could tip the scales in favor of plaintiffs. Additionally, it exposes the lack of transparency among AI startups, which often hide their data sources under trade secrets. According to a 2024 RIAA report, generative AI music could cost the industry up to $2 billion in lost royalties by 2028 if not properly regulated.

Immediate Consequences

• Lawsuits and Regulation: We are likely to see class-action lawsuits from musicians and record labels. The European Union is already preparing the AI Act, which would require transparency in training data. In the US, Congress has introduced the 'Generative AI Copyright Act' that would force companies to disclose their data sources and compensate rights holders.
• Change in Practices: Companies like OpenAI have begun signing agreements with content publishers, such as the pact with Axel Springer in 2023. Suno may be forced to license its data or redesign its model, which would involve an estimated cost of tens of millions of dollars in new data acquisitions.
• Impact on Users: If Suno is forced to remove its current model, users could lose access to the tool or face changes in the quality of generated songs. The platform has over 10 million monthly active users, according to 2025 data, many of whom use Suno to create music for free or through subscriptions.

What Readers Should Know

Not all use of scraped data is illegal; it depends on each country's laws and whether the content is copyrighted. However, the leak shows that Suno did not have licenses for most of the works. For creators, this reinforces the need to protect their works with metadata and monitor AI platforms. For users, it is a reminder that free tools are often built on infringements. Moreover, the Suno case could set a legal precedent: if courts determine that mass scraping for AI training is not 'fair use,' many startups could collapse.

“The Suno leak is a moment of truth for generative music: either it gets regulated or it becomes unsustainable,” comments an analyst from TheVortiq.

Technical Analysis

The leaked code shows that Suno used massive web scraping techniques, including downloading audio and text files. The dataset is estimated to contain over 50 million songs. This contrasts with companies like Anthropic, which have chosen to train on public domain or licensed data. Suno's lack of transparency now places it in a vulnerable position before regulators and courts. The code also reveals that Suno employed audio recognition models to transcribe lyrics and metadata, suggesting a sophisticated extraction process that may have violated the terms of service of platforms like Spotify and YouTube.

Looking Ahead

This incident will accelerate the implementation of legal frameworks for generative AI. Countries like Japan have already established exceptions for data mining, but in the US and Europe, the trend is toward greater copyright protection. Suno could serve as a landmark case defining the limits of using artistic works to train AI. Additionally, the leak has caused investors to reconsider their bets on generative AI startups: according to PitchBook, funding in this sector fell 15% in Q3 2025 compared to the same period last year. For users, Suno's future is uncertain: it could survive through licensing agreements, as Spotify has done with record labels, or disappear if it fails to adapt to the new regulatory environment.