Millions of books died so Claude could live

Summary of Millions of books died so Claude could live

by The Verge

1h 28m•February 3, 2026
Listen to Episode

Overview of The Vergecast — "Millions of books died so Claude could live"

This episode of The Vergecast (host David Pierce) covers three main topics: a deep-dive into Anthropic’s Project Panama and how publishers’ books were acquired and used to train large language models (guest: Will Oremus, Washington Post), Netflix’s evolving relationship with theatrical movies and what that means for the future of cinemas (guest: Julia Alexander), and a VergeCast hotline/smart‑home troubleshooting segment focused on IKEA’s new Matter-over-Thread devices and why inexpensive smart buttons aren’t working for many users (Jen Toohey).

Key segments

1) Project Panama, books, and AI (Will Oremus)

  • What Project Panama was

    • Anthropic attempted a large-scale project to digitize books to feed into its Claude models; court documents described the goal as to “destructively scan all the books in the world” (they sliced spines to speed scanning — the often-cited “hydraulic-powered cutting machine”).
    • Books were attractive because, on average, they are higher-quality, edited, and potentially contain information not easily scraped from the open web — useful to “bootstrap” model quality.

  • How the data was acquired

    • Early steps involved downloading existing pirated, digitized “shadow libraries” (e.g., LibGen); some OpenAI staff reportedly did similar downloads in earlier years.
    • For large-scale physical acquisition, Anthropic employed people experienced in mass-scanning projects (e.g., ex–Google Books personnel), bought vast lots of used books from warehouses and sellers, and contracted fast, destructive scanning vendors. Orders mentioned in filings ran into the hundreds of thousands of books.

  • Legal and ethical outcomes

    • Multiple lawsuits (authors, publishers) have targeted how AI companies obtained and used copyrighted works.
    • Two judges (in different cases) ruled that training on book content could be fair use — but for different reasons and with differing analyses (one called such use “exceedingly transformative,” another left open the possibility it might hurt markets).
    • Where Anthropic ran into trouble in court was how it acquired and stored some books — especially those purchased or held but not used commercially — which led to significant legal exposure. Anthropic settled some claims (reported $1.5 billion settlement).
    • Overall copyright law in this area is unsettled: early rulings favor some forms of training as fair use, but the decisions vary and appeals / future rulings could change the landscape.

  • Broader implications and the “original sin” theory

    • David’s framing: the industry’s current legal/ethical backlash partly stems from OpenAI’s early, cavalier sourcing practices that set a precedent and started a race where taking shortcuts felt necessary to compete.
    • Anthropic and others faced a moral paradox: the belief that “losing the AI race” justifies aggressive data acquisition. That mentality — and the scale of commercial stakes — fueled practices that creators see as exploitation.
    • Implication: this controversy fuels calls for regulation, clearer licensing/compensation models, and new industry norms around data sourcing.

  • Notable imageable details

    • “Destructively scan” (one book copy was destroyed per title for speed).
    • Shadow libraries (LibGen) as initial shortcuts.
    • Hiring experienced mass‑scanning personnel (e.g., ex–Google Books) to scale physical acquisition.

2) Netflix, theatrical movies, and the future of movie theaters (Julia Alexander)

  • Netflix’s theatre posture

    • If Netflix acquires Warner Bros. Discovery, it could become a top theatrical player — but Netflix’s public messaging has fluctuated (historically de-emphasized theaters; now promises to support them).
    • Netflix argues theatrical releases can boost streaming engagement, but the company lacks traditional theatrical distribution infrastructure and may only give selective theatrical windows for prestige/award titles or IMAX runs.

  • Why theaters still matter

    • Perception of quality: theatrical releases signal prestige and event-ness; audiences still associate theaters with “bigger” experiences.
    • Marketing and fandom: big theatrical releases build awareness and social momentum that can boost downstream streaming viewership.

  • Industry realities and constraints

    • IMAX and premium screens are scarce; there is fierce competition for those slots.
    • Studio economics: theatrical blockbusters are expensive and risky; the middle-budget film (the old “$30–$50M” sweet spot) is diminishing.
    • Box office recovery has been uneven post‑pandemic; supply of theatrical films has decreased, impacting exhibitors’ ability to fill calendars.

  • Possible paths for theaters

    • Eventization: make theaters a place for communal, “social capital” experiences (e.g., Barbenheimer-style cultural moments, horror fandom, anime).
    • Re-releases & nostalgia: reprints of beloved films can fill exhibitor calendars affordably and bring audiences out.
    • Alternative content: curated events, anime, live broadcasts, and even short-form/viral content marathons (TikTok/YouTube shorts nights) are plausible experiments.
    • Designated phone-friendly/rowdy screenings: explicitly marketed screenings where phone use or interactive behavior is allowed, preserving other screenings for classic quiet experiences.

  • Concerns

    • Netflix “Netflixifying” filmmaking: studios asking creators to design films for streaming attention spans (early big set pieces, repetitive plot cues) could further erode theatrical incentive and cinematic craft.
    • Question of stewardship: if Netflix acquires studios, will it preserve theatrical traditions or repurpose them for streaming-first strategies?

3) VergeCast hotline — IKEA Matter devices and smart‑home headaches (Jen Toohey)

  • The recent IKEA rollout

    • IKEA introduced a very low-cost line of Matter-over-Thread devices, including inexpensive physical buttons (~$6), sensors, bulbs, etc. Big potential: mass-market, affordable Thread/Matter devices.

  • The main problem callers reported

    • Many users cannot get IKEA’s buttons to work with Google Home. Buttons are supported inconsistently across platforms.
    • Google Home (as of the episode) did not support Matter “button” / generic switch device types — a major platform-level gap.
    • Other platforms that do support buttons: Amazon Alexa, Apple Home, Home Assistant, SmartThings.

  • Thread/network complexity

    • Thread devices require a Thread border router. Multiple border routers and fragmented implementations (Eero, Amazon, Apple, IKEA) can create incompatible Thread networks.
    • Devices can pair OK initially, then drop off or fail to trigger routines (common pain points).
    • Finding/entering a Thread network key and troubleshooting cross‑vendor issues is confusing for many consumers.

  • Practical advice & troubleshooting

    • If you need reliable IKEA buttons now: avoid Google Home (it lacks button support); prefer platforms that currently implement button support (Alexa, Apple Home, Home Assistant).
    • Aim to have a single, consistent Thread border router (to avoid fragmented Thread networks); Home Assistant can be useful for advanced troubleshooting.
    • IKEA has good returns — try exchanging defective hardware if pairing fails.
    • Expect bugs and iteration: this is one of the first large, mass-market Thread/Matter rollouts; problems are likely to be addressed via firmware and platform updates.

Main takeaways / recommendations

  • For creators & publishers:

    • The legal picture around AI training and copyright is unsettled. Monitor ongoing cases and consider negotiating licensing models or advocating for clearer rules/regulation.
    • The history of data sourcing matters: how models were built is shaping public and legal reaction; transparency and proactive licensing could reduce backlash.

  • For consumers interested in smart homes:

    • Don’t assume Matter means “plug-and-play” across every ecosystem yet. Check platform compatibility (Google Home currently lags on buttons).
    • If you want IKEA’s cheap buttons now, use a compatible platform (Alexa, Apple Home, Home Assistant) and ensure your Thread border router configuration is correct.
    • Expect some growing pains as Thread/Matter scales; IKEA’s pricing and scale are promising for broader adoption.

  • For moviegoers and the industry:

    • The theatrical experience still has unique value (communal, event-driven, premium formats). Exhibitors should lean into experiences and fandom-driven events.
    • Re-releases, curated programming, and alternative content windows are practical ways to fill calendars and restore foot traffic.
    • If you care about diverse cinema (mid-budget original films, auteur projects), advocate for distribution choices and support theatrical event screenings when they appear.

Notable quotes / memorable lines

  • “Project Panama … to destructively scan all the books in the world.” — description from reporting on Anthropic’s project.
  • “Hydraulic-powered cutting machine” — vivid phrase used to describe book spine removal for high-speed scanning.
  • Judge characterization (quoted): training on books can be “exceedingly transformative” — but judges disagree on scope and market effects.
  • David’s “original sin” theory: OpenAI’s early sourcing decisions set a precedent that accelerated a race where many actors felt shortcuts were necessary.

Things to watch (next steps / follow‑ups)

  • Further court rulings and appeals on AI training and fair use — could reset industry practices.
  • Netflix’s behavior if it acquires Warner Bros. Discovery: will it commit to wide theatrical slates or keep selective, event-first windows?
  • IKEA’s April product wave and firmware/platform updates: will initial pairing/Thread issues be resolved at scale?
  • Google Home platform updates: will Google add Matter button support and improve Matter coverage across device types?

If you want a concise bullet summary of just legal outcomes, or a short “what to do if your IKEA button doesn’t work” checklist, say which you’d prefer and I’ll produce it.