3JUN

Alsup withholds approval on Anthropic $1.5bn

4 min read

08:55UTC

Judge William Alsup declined to grant final approval of the $1.5bn Bartz v. Anthropic class settlement at the Thursday 14 May hearing, asking for more detail on attorneys' fees and lead-plaintiff payments. The figure is a class settlement floor, not the bilateral News Corp deal U#1 reported.

← Back to Lenovo runs football's biggest AI broadcast Jump to analysis ↓

IndustryDeveloping

Key takeaway

Alsup withheld approval of the $1.5bn Bartz settlement on 14 May, reframing News Corp's figure as a litigation floor.

Judge William Alsup held the final-approval hearing for Bartz v. Anthropic in the Northern District of California on Thursday 14 May and declined to grant approval, instead requesting more detail on attorneys' fees and lead-plaintiff payments ¹. Anthropic has already paid $300m into the settlement fund after Alsup's preliminary approval in September 2025; a further $300m is due within five days of final approval, with the balance phased thereafter. Bartz certifies authors and publishers whose work Anthropic ingested without licence as the affected class.

The hearing forces a correction to Lowdown's U#1 lead. Lowdown U#1 reported Robert Thomson's Q3 FY26 disclosure of an anticipated $1.5bn Anthropic settlement as one of three publisher revenue templates, alongside Reach plc's pay-per-usage deal with AWS and the NYT-Amazon flat-fee arrangement. That framing read the $1.5bn as a bilateral price; the docket reads it as News Corp's prospective share of the Bartz class-action settlement, not a bilateral News Corp-Anthropic licensing deal. Thomson's quoted phrase was "anticipated" rather than "agreed"; read against the settlement timetable, that wording is accurate, but read as a bilateral term sheet, it is not.

The distinction sits at a different layer of the publisher-AI market. The live commercial templates a mid-tier publisher actually has to choose between today are these: Reach plc's pay-per-usage arrangement with AWS ; DotDash Meredith's reported $16m flat-fee licence with OpenAI; and the New York Times flat-fee deal with Amazon at around $20m, structurally similar to DotDash. Each is a per-counterparty deal with disclosed terms. Bartz operates at a different layer entirely: a class settlement covering what unlicensed scraping costs industry-wide, not a ceiling on what licensed access can earn. News Corp's existing bilateral AI licensing income remains the previously reported $250m five-year OpenAI deal.

The correction does not soften U#1's editorial through-line: the publisher-AI market is being re-priced quarter on quarter, and Thomson chose to read a settlement number into an SEC-filed transcript as a negotiating instrument. It does shift the locus from commercial template to litigation floor. If Alsup restructures the deal on attorneys-fees grounds at the reschedule, every plaintiff-side AI copyright complaint will file for similar fee adjustments within the quarter, and Reach-style pay-per-usage rate cards will re-price upward against a confirmed unlicensed-scraping floor.

Deep Analysis

In plain English

Anthropic is the company behind the Claude AI. A group of writers and publishers sued Anthropic, claiming it used their published work to train its AI without permission or payment. They reached a $1.5bn settlement. A federal judge in San Francisco was reviewing that settlement on 14 May and decided not to approve it yet. He wants more information about how much the lawyers will be paid before he signs off. Anthropic has already put $300m into the fund; another $300m would follow once the judge approves.

Deep Analysis

Root Causes

The Bartz settlement reached this stage through two structural conditions. First, Anthropic's training data ingestion without licensing was not an inadvertent oversight; the company's 73,000-pages-per-human-visitor scrape ratio (documented by the Center for Journalism & Liberty) reflects a deliberate scaling choice that maximised training data at the expense of publisher licensing costs.

Second, the class action mechanism was the only viable vehicle for publishers who individually lacked the litigation resources to sue Anthropic at the scale necessary to move the company.

News Corp's use of an SEC-filed earnings transcript to disclose the anticipated $1.5bn figure was a deliberate information-asymmetry play: by reading a class-settlement share into a formal investor channel as if it were a bilateral deal, Thomson reset the bilateral licensing benchmark against which every other publisher was negotiating.

The correction this event forces does not undo that negotiating effect; it does clarify the legal architecture for publishers choosing between litigation and direct deals. The $1.5bn is a class-settlement share, not a bilateral price.

What could happen next?

Meaning
If Alsup converts his fee query into a class-adequacy challenge, the $1.5bn figure becomes reopenable, which would raise Anthropic's total liability exposure and accelerate every other pending AI copyright plaintiff toward settlement. The distinction between class-settlement floor and bilateral licensing ceiling is now live in the public record, which means publishers in active bilateral negotiations with Anthropic can cite the settlement as a floor while keeping the bilateral ceiling open.

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: Judge Denny Chin's 2011 rejection of the Google Books settlement after the Authors Guild v. Google class action had run for six years provides the structural template. Chin declined final approval on adequacy-of-representation grounds: the settlement gave Google a de facto monopoly on orphan works without Congress authorising it.

Alsup's intervention is less sweeping: he is seeking fee clarification rather than questioning the settlement's market structure implications. The underlying dynamic is the same: a transformative-technology copyright settlement that the court must approve as fair to a class it cannot fully survey.

Counter-parallel: The 2014 settlement in Authors Guild v. HathiTrust, where Judge Baer granted final approval after a single round of supplemental briefing on fees, shows the more common trajectory: fee-detail requests are resolved in one additional hearing without disturbing the underlying agreement.

The $300m Anthropic has already paid into the settlement fund is committed regardless of Alsup's final approval timing. The next $300m tranche, due within five days of final approval, is the mechanism Alsup is controlling. Every month of delay costs Anthropic the opportunity cost of that $300m at its current annualised growth rate (80x per Q1 2026 per Dario Amodei), though the absolute dollar cost of the delay is marginal against Akamai's $1.8bn seven-year commitment signed the same week.

If Alsup restructures the settlement on adequacy grounds, every pending plaintiff-side AI copyright complaint will file for similar structural review, resetting publisher pay-per-usage rate cards against a higher litigation floor. Reach-style pay-per-usage rate cards will re-price against a confirmed litigation floor that is higher than the current bilateral deal benchmarks (DotDash Meredith at $16m per year, New York Times at approximately $20m per year from Amazon).

Consensus view: The Tow Center for Digital Journalism's research on AI copyright litigation has found that fee-structure challenges in class actions of this size typically extend timelines by three to nine months but do not reverse preliminary approvals.

Alsup's specific request for more detail on attorneys' fees and lead-plaintiff payments is procedurally standard for a settlement at this dollar value; the Ninth Circuit requires detailed lodestar cross-checks on any plaintiff-side fee request above $100m.

Counter-view: Electronic Frontier Foundation litigation counsel has argued that Alsup's interventions in AI copyright cases have been structurally sceptical beyond procedural review, citing his management of the GitHub Copilot class action in the same court.

His track record suggests his fee query may also be probing whether the class is adequately represented given that News Corp's anticipated $1.5bn share dwarfs the lead plaintiff's stake, which could reopen class-certification questions.

Key tension: Whether Alsup uses the fee-detail hearing as an entry point to reassess the adequacy of class representation, which would extend the timeline significantly and give Anthropic grounds to renegotiate the total figure before final close.

Sources:Adweek

Mentions:Judge William Alsup →Anthropic →News Corp →Robert Thomson →Google →California →San Francisco →DotDash Meredith →OpenAI →Reach plc →Amazon →New York Times →Claude Cowork →

First Reported In

Update #2 · BuzzFeed sold for AI; agentic ads go primetime

Adweek· 17 May 2026

Read original →

Causes and effects

Caused by

News Corp names $1.5bn Anthropic settlement

Bartz v. Anthropic correction directly responds to U#1's framing of Thomson's $1.5bn 'anticipated settlement' figure as a bilateral licensing template.

Occurred 8 May 2026

Read story →

Reach signs AWS pay-per-usage AI licensing deal

Bartz settlement floor clarifies the distinction between bilateral licensing deals (Reach/AWS, NYT/Amazon) and litigation-derived class-action floors.

Occurred 1 Apr 2026

Read story →

This Event

Alsup withholds approval on Anthropic $1.5bn

Corrects U#1's framing of Robert Thomson's $1.5bn 'anticipated' figure and reclassifies it as the litigation floor on unlicensed scraping liability, structurally distinct from the live bilateral publisher templates.

Led to

Judge probes $1.5bn Anthropic class deal

Judge Alsup's prior declination of the Bartz settlement on fee grounds set the procedural context in which Judge Martínez-Olguín (now reassigned) is conducting fresh scrutiny of the settlement terms.

Occurred 18 May 2026

Read story →

Different Perspectives

DAZN

DAZN inherits Lenovo's entire FIFA World Cup 2026 AI broadcast stack without a procurement decision, repeating the vendor-dependency pattern from its TikTok Serie A livestream. Its own Delta Protocol moved to a funded IBC 2026 Accelerator proof-of-concept in the same week, the build-side response to that structural exposure.

European public broadcasters and regulators

Canal Alpha's Harmonic EyeQ deployment and Meta One's non-EU market sequencing both land with 60 days until the EU AI Act Article 50 transparency obligations take effect on 2 August. No major EU broadcaster has signed the Code of Practice; regional public broadcasters face a compressing window to disclose AI-encoding postures that may fall within or outside synthetic-content obligations.

Jakob Owens and the creator community

Artlist's launch of an all-AI streaming channel on $300m of recurring revenue from the very filmmakers it serves drew the week's sharpest reaction: Owens said publicly he hoped it fails. The backlash is the signal that creator-tool vendors crossing into self-distribution face a customer-conflict dynamic that Getty Images' 2021 Unsplash play already documented.

Artlist

Artlist launched Artlist TV, an all-AI streaming channel built on its own tools, alongside disclosing $300m in annual recurring revenue, crossing from selling creative tools into self-distributing finished shows. That move puts the Israeli vendor in direct competition for attention with the 50 million filmmakers and creators who pay for its products.

Lenovo

Lenovo's 2 June World Cup announcement is its most direct public claim to being a media-layer company rather than a hardware supplier: the 17,000-device, three-AI-layer deployment positions it as the entity that sets the broadcast standard FIFA's rights-holders inherit. The Official Technology Partner designation converts marketing spend into structural control of the largest live AI broadcast ever staged.

ByteDance / CapCut

ByteDance accepted Google Gemini orchestration for CapCut despite US-China platform sensitivity in 2026, confirming that orchestration-layer user demand overrides geopolitical friction at the product level. TikTok simultaneously faces the DDEX labelling pressure from Spotify's standard.