↓Skip to main content
Zeitgeist — a spike by Chris Gathercole
  1. Topics/
  2. Data, IP & Training Rights/

Gather — 23 July 2026

Bartz v. Anthropic Reaches Final Approval #

  • Anthropic’s landmark $1.5B copyright settlement is approved (TechCrunch, 2026-07-20/21) — U.S. District Judge Araceli Martinez-Olguin granted final approval July 21, closing out the largest copyright settlement in US history — first tracked in this journal since May 19 as a preliminary settlement. 91%+ of eligible authors/publishers have claimed; $101M in attorneys’ fees approved (down from a requested $187.5M). Authors who opted out retain separate suits. This converts five gathers’ worth of “pending settlement” tracking into a closed case — the reference point other AI-copyright negotiations (Meta, Google, Suno/Udio) will now cite as settled precedent.

Music Litigation: NY Denies Expansion, Sony Immediately Refiles #

Publisher/Platform Litigation: NYT Sharpens Its Microsoft Theory #

  • New York Times Trims OpenAI Suit, Targets Microsoft’s Conduct (Bloomberg Law, amended complaint filed 2026-06-25) — NYT’s amended complaint newly alleges Microsoft actively enabled OpenAI’s infringement by building a custom supercomputing system specifically to make large-scale training on NYT content possible — shifting part of the theory from “OpenAI trained on our content” to “Microsoft built the infrastructure knowing what it would be used for.” Two claims (trademark dilution, contributory infringement) were dropped in the same amendment.

Data Licensing Expands to a New Sector: Biopharma #

  • AI data becomes sticking point in biopharma dealmaking (BioSpace, 2026-07-22) — Data ownership/access terms are now the primary friction point in pharma–AI-biotech partnerships (Merck–Protillion $510M pact, June 2026; multiple Lilly/BMS/Incyte deals, May 2026) as AI drug-discovery platforms need continuous data access rather than one-time licenses to keep improving. The first sector-specific data-licensing dynamic tracked here outside media/publishing/music — the “real-time access, not one-time acquisition” shift documented for content licensing (2026-06-26 gather) is recurring independently in life sciences.

State Legislation: California’s AB 412 Shelved #

  • California’s AB 412 Still Demands Developers Do The Impossible (EFF, 2026-06) — AB 412 (would require AI developers to publish a mechanism for rights-holders to query use of their copyrighted training material) was designated a two-year bill by the Senate after passing committee 6-2 — effectively shelved until 2027. EFF’s critique: identifying specific copyrighted works used in training at the granularity the bill demands is not technically achievable with current methods.
  • AI Impact on Society The Bartz settlement’s final approval is a settlement-lifecycle precedent (filed → preliminary → final → distribution) other AI-copyright and societal-impact coverage will reference.
  • Claude Integrations NYT v. Microsoft’s amended complaint bears on the Microsoft/OpenAI infrastructure relationship, relevant to how tightly-coupled cloud/model partnerships are scrutinized.
  • Vibe Coding Approaches xAI’s Grok Build open-sourcing followed a privacy scandal in which the tool silently uploaded user directories (including SSH keys/passwords) to xAI’s cloud — a data-handling incident distinct from the training-data IP concerns tracked here, but worth a future entry if litigation follows.

Meta-observations #

  • Emerging pattern: Refile-after-denial (Sony v. Udio) is a new litigation tactic worth watching — plaintiffs facing an adverse scope-expansion ruling in one suit are filing a parallel second suit naming the excluded works rather than appealing. Watch for this in the Boston Suno case if Judge Saylor denies the 61,026-works expansion there too.
  • Quality signal: TechCrunch’s Bartz final-approval coverage is the first primary-outlet confirmation that a major AI-copyright settlement has actually closed (not just been preliminarily approved) — worth treating as the template for how this journal tracks “settlement lifecycle” going forward.
  • Emerging theme: Real-time/continuous data-access licensing (as opposed to one-time training-data acquisition) is recurring across independent sectors — content/media (Pebblous, 2026-06-26) and now biopharma (BioSpace) — suggesting a structural shift in how AI companies contract for data generally, not a media-industry-specific phenomenon.
  • Keyword suggestion: "AI data licensing" biopharma OR "life sciences" 2026 — biopharma AI-data dealmaking is a distinct beat from the media/publisher licensing coverage this journal’s current keywords surface.
  • Gap: No primary-source (docket-level) confirmation yet of Sony’s July 20 second Udio suit beyond trade-press coverage (MusicNews) — worth verifying against a litigation tracker next cycle.