The man who fought Napster is now fighting AI

Host Nilay Patel explains that Napster pushed him toward copyright law, where he later defended students sued for using Kazaa, while Mitch Glazier was a young RIAA lawyer on the opposing side. Glazier recalls being sent to college campuses to argue against peer-to-peer file sharing, including one hostile appearance where he walked onstage with a garbage bag over his head because students wore anti-RIAA shirts and he expected tomatoes. He says those conversations nevertheless taught him about fan-driven technological change and the disconnect between industries and the people reshaping them.

Patel says his views have shifted from celebrating open internet remixing to worrying that AI platforms extract the economic value of art while flooding culture with cheap material. He asks how to protect creators without making new creation harder. Glazier says “balance” is the right goal: unlike during Napster, artists, fans, labels, and newer artist-service companies are broadly rowing in the same direction. He attributes that alignment to a more diverse music economy and says the central issue is preserving an authentic human connection between artist and fan, not merely defending an old business model.

Glazier argues that copyright exists to give people enough control and income to keep making art, rather than forcing artistic work to remain a hobby. He worries that music generated in seconds without lived experience could dilute the incentives for human musicians and move culture backward even as technology advances. Patel contrasts CDs, whose scarcity created a clear price, with streaming, where copying became frictionless, and AI, where creation may become nearly frictionless. Glazier says the new scarcity will ultimately be determined by fans, who value an artist’s persona, brand, and emotional relationship as well as the recording.

Glazier distinguishes fully generative AI from assistive AI. He expects fans to continue rejecting music made entirely by a machine because it lacks a human connection, but says AI used as a tool could eventually become as ordinary as synthesizers or Auto-Tune. He does not consider Patel’s example—typing prompts into Suno, refining the output, and uploading it through DistroKid—to constitute the same creative relationship. His policy priority is securing creators’ basic rights first; once rights are protected, labeling and consumer choice can determine which business models survive.

Glazier describes the RIAA as a 46-person Washington organization that began as a technical standards body, creating the equalization curve that made vinyl records playable across equipment. It still works on international metadata standards and runs the Gold and Platinum certification program, publishes recorded-music revenue data twice a year, represents major labels when the industry is aligned, builds coalitions with independent artists and companies, and operates a 24-hour content-protection operation. That operation handles takedown requests for leaked releases and works with sister organizations in London and New Zealand. Glazier says litigation and lobbying are important, but not the whole organization.

Glazier says he avoids surrounding himself with “yes people,” seeks raw opinions from his team, checks his instincts with trusted contacts, and tries to understand how opponents would argue before judges or policymakers. About 60% of the time he keeps his original gut reaction; about 40% of the time outside expertise changes his path. For major cases, such as a lawsuit that could define creators’ AI rights, he wants broad internal and external input. He also distinguishes between pursuing a practical solution and deliberately making “good trouble”: the former requires a realistic path to a landing zone and some moderation inside his own industry.

Patel asks who could object when Instagram fills with videos placing celebrities, politicians, or fictional characters into another artist’s song: the performer, the character owner, the AI company, the uploader, or the platform. Glazier says U.S. law provides no general federal right to control one’s voice or likeness, while state publicity rights usually focus on advertising and endorsements. The proposed federal No Fakes Act would create a new intellectual-property right closer to trademark than copyright. It would combine hurdles against false takedown notices, counter-notices, and rapid judicial review so platforms would not decide constitutional questions involving parody, news, bullying, or other protected speech.

Patel contrasts YouTube’s licensing relationship with music against Google Search’s repeated fair-use defenses for indexing images and books, arguing that Google’s AI division may be treating training data as fair use while YouTube understands that music is essential to its business. Glazier says YouTube learned that winning an intellectual-property case could still mean losing the fan experience, which led it to develop Content ID and a deeper licensing partnership. He says the RIAA’s case against Suno argues that training on music is not fair use because generated music substitutes for human-made music in the same marketplace, unlike a search index that can direct users toward the original work.

Glazier says the RIAA is pursuing policy and litigation to establish that companies cannot reproduce music for training without consent or compensation. He points to Anthropic’s IPO filing, whose risk disclosures acknowledge possible liability for scraping, as evidence that AI companies understand the legal exposure. Patel argues that artists now have more economic stake than during Napster because streaming makes catalogs evergreen and lets musicians sell catalogs to investors such as Goldman Sachs. Glazier says licensed deals with companies such as Suno or Udio could include limits, walled gardens, or download restrictions that prevent AI music from overwhelming Spotify and other services, although he is not privy to those contracts.

Patel questions whether Spotify will label AI music when it has financial incentives to reduce payments for human recordings and already struggles with fake streams and low-listenership uploads. Glazier says Spotify has begun labeling some AI information only in credits, which is insufficient, and argues that disclosure must happen at the track level. He expects both policy and competition to push platforms further: laws such as California’s transparency rules can force global implementation, while Apple and Spotify may imitate whichever competitor appears more subscriber-friendly. For fully generated music, Glazier says detection technology works well with low false positives, but distributors such as DistroKid and TuneCore—not only record labels or streaming platforms—must identify and tag uploads.

Patel asks how awards and charts can count music when listeners discover fragments on TikTok, AI remixes spread socially, and Billboard methodology remains contested. Glazier says the RIAA still uses the equivalent of 1,500 streams per unit and can certify albums with reasonable certainty, but chart systems are beginning to require music to be “substantially human.” Billboard currently permits some AI material with labeling, while other chart organizations exclude it from eligibility. Glazier expects human streams to remain part of certification and says the harder question will be how to evaluate assistive AI. The standard may evolve, but the creator must remain meaningfully in control; entering a few words into a generator is not equivalent to human authorship.