Numerous music industry groups and the three majors \u2013 Sony Music Entertainment, Universal Music Group, and Warner Music Group \u2013 have thrown their support behind a proposed new US law that would require AI developers to disclose the materials they used to train their AI models.<\/p>\n
The Transparency and Responsibility for Artificial Intelligence Networks<\/strong> (TRAIN<\/strong>) Act<\/strong>, introduced by Sen. Peter Welch<\/strong>, a Vermont Democrat, would apply only when a rightsholder suspects their works were used to train a generative AI tool.<\/p>\n The rightsholder would be able to ask any US district court clerk to issue an administrative subpoena to an AI developer, requiring them to hand over materials that are \u201csufficient to identify with certainty\u201d whether their copyrighted works were used.<\/p>\n A subpoena would be granted only if the rightsholder declares that they have \u201ca good faith belief\u201d that their work was used to train the model.<\/p>\n Sen. Welch said in a statement on Monday (November 25) that the bill is meant to address the \u201cblack box\u201d problem with AI developers\u2019 use of copyrighted materials. Simply put, AI developers often don\u2019t reveal what data or data sets they used to train their AI.<\/p>\n This makes it difficult for rightsholders to know if their works have been used in AI training without their authorization, and it puts the onus on rightsholders to prove their material was used by an AI developer when they take legal action against that developer.<\/p>\n For instance, in the cases against chatbot developer Anthropic<\/strong><\/a>, and generative AI music apps Suno<\/strong> and Udio<\/strong><\/a>, music companies have gone to great lengths to show similarities between the AI output and their copyrighted materials. The TRAIN Act would simplify and speed up that process.<\/p>\n \u201cThis is simple: if your work is used to train AI, there should be a way for you, the copyright holder, to determine that it\u2019s been used by a training model, and you should get compensated if it was,\u201d Sen. Welch said.<\/p>\n \u201cWe need to give America\u2019s musicians, artists, and creators a tool to find out when AI companies are using their work to train models without artists\u2019 permission. As AI evolves and gets more embedded into our daily lives, we need to set a higher standard for transparency.\u201d<\/p>\n “If your work is used to train AI, there should be a way for you, the copyright holder, to determine that it\u2019s been used by a training model, and you should get compensated if it was.”<\/p>\n Sen. Peter Welch (D-VT)<\/span><\/p><\/blockquote>\n Notably, the TRAIN Act doesn\u2019t require AI developers to compensate copyright holders for the use of their works in training AI, only to disclose such use.<\/p>\n A separate bill \u2013 the Content Origin Protection and Integrity from Edited and Deepfaked Media Act<\/strong> (COPIED Act<\/strong>) \u2013 that was introduced in the Senate<\/a> earlier this year would make it unlawful to use copyrighted works to train AI without permission \u2013 a key ask of many music industry organizations and companies.<\/p>\n In copyright cases before the courts, AI companies have argued, among other things, that using copyrighted works to train AI should benefit from the \u201cfair use\u201d exemption to copyright law. The copyright holders suing AI developers strenuously object to that, arguing that fair use was never meant to be applied to situations like the mass ingestion of copyrighted materials to train generative AI tools that then regurgitate similar or identical content.<\/p>\n Courts are considering whether or not the use of copyrighted works in AI is \u201cfair use\u201d in several copyright lawsuits.<\/p>\n The TRAIN Act and COPIED Act are among a growing number of bills before Congress aimed at regulating the use of AI. Others include the NO FAKES Act<\/strong><\/a>, introduced in the Senate this year, which would allow individuals to sue if their voice or likeness is used in an AI deepfake. A similar bill, the No AI FRAUD Act<\/strong><\/a>, was also introduced in the House of Representatives this year.<\/p>\n The TRAIN Act is similar in intent to \u2013 but different in method from \u2013 the Generative AI Copyright Disclosure Act, which Rep. Adam Schiff<\/strong>, a California Democrat, introduced earlier this year in the House of Representatives.<\/p>\n That bill also seeks transparency from AI developers in the materials they use, but it does so by requiring AI companies to send a notice to the Register of Copyrights that includes \u201ca sufficiently detailed summary of any copyrighted works used.\u201d<\/p>\n \u201cWe strongly support the bill which prioritizes creators who continue to be exploited by unjust AI practices.\u201d<\/p>\n