The International Confederation of Music Publishers (ICMP) has launched an investigation alleging that AI companies are “illegally” using copyright-protected music to train their AI models.
According to a report shared with Billboard, the ICMP’s investigation has compiled over two years of evidence gathered from public registries, open-source training content repositories, leaked materials, and AI research studies. The investigation claims that AI firms are training their models using music from artists such as The Beatles, Beyoncé, The Weeknd, Gorillaz, and Kanye West without obtaining the necessary licenses.
The ICMP alleges that AI tools, including Google’s Gemini, Anthropic’s Claude, Microsoft’s CoPilot, and Meta’s Llama 3, have “scraped” music from platforms like YouTube to train their models. The ICMP asserts that its evidence is “comprehensive and clear,” suggesting that the scale of digital music being used to train AI is “larger than previously acknowledged.”
The investigation also singles out X’s built-in AI chatbot, Grok, as a major offender, alleging that it copies and distributes copyrighted lyrics. ICMP Director John Phelan characterized the situation as “the largest IP theft in human history,” stating that “tens of millions of works” are being infringed upon daily.
Phelan further explained that many companies are scraping lyric datasets from the internet, incorporating millions of works into their models, which he claims is a “direct breach” of copyright laws and regulations.
The full investigation report is available via Billboard.
