← All media signals
Governance

The Training Consent Wars

Writers, artists and coders fight over what fed the machine, demanding a say in the work that became its raw material.

What is it?

Writers, artists and coders are contesting what trained the model, in court and at the table, as settlements set a price on human work used as raw material. Anthropic’s $1.5B is the largest copyright settlement yet. Whether that price generalises, or was bought by circumstance, stays open.

Why now?

Regulation always lags the technology it governs, and training data is where that lag turned litigable. Every model already shipped was trained before consent had a price, so today’s claims are retroactive by design. The bill comes due after the product ships, never before it.

Who’s building it?

AI labs, publishers, and creator collectives negotiate the terms, but courts are setting the actual price first.

AI labsPublishers & rightsholdersCreator collectivesCourts & regulatorsLicensing platforms
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — AI labs, and licensing platforms rising to serve them.
Buyers — publishers, musicians, and writers, deciding whether to sue or to sign.
Gatekeepers — courts, and the settlements they approve, setting the market’s opening price.
What’s at stake
Every open case sets a price the next negotiation starts from — a settlement anywhere raises the floor everywhere.
Reputational exposure for firms still relying on unlicensed data.
Whether creators gain leverage, or just a one-time payout.
What’s unsettled
Whether the settlement price becomes a floor, or stays a ceiling.
Whether text and code follow music’s settlement pattern, or diverge.
Whether licensing becomes standard practice, or a defensive move made only once sued.
02 · How the demand matureda progression, not a timeline
Litigation
Rightsholders sue first, forcing courts to name a price for the harm.
Settlement
the price goes public, and every other lab must reckon with it.
Licensing rails
publishers and platforms build standing deals so future training doesn’t need a lawsuit first.
Standard
per-use pricing becomes the default, and the fights move to what’s left unsettled.
03 · Where the market goesthe forks, not a prediction
→Licensing or litigation — the market may settle into standard per-use rates, or stay a running docket of individual court fights.
→Rightsholders could organise collectively for leverage, or negotiate alone and take whatever terms are offered.
→New models could license data upfront, sidestepping the fight the earlier generation still pays for.
→Or courts rule scraping was fair use, and the settlements stay exceptions.
Everything here turns on precedent: whether a court-ordered price for one company’s infringement becomes the market’s price for every model built after it.
04 · The Compliance-as-Feature deskadjacent markets, moving now
The Model Union
Creators unionising to negotiate collective payment
Anthropic pays about $3,100 per work, across 480,000+ works
TechCrunch
Consent Fatigue
Platforms quietly repurpose user data to train models
LinkedIn began using member data to train AI models
LinkedIn
The Liability Question
Who answers when an AI system causes harm
US court let Character.AI product-liability claims proceed in 2025
National Constitution Center
05 · The Wire
01
Anthropic’s $1.5B author settlement won final approval
TechCrunch ↗
02
Music publishers sue Anthropic over $3B+ in claimed damages
TechCrunch ↗
03
Warner Music settled with Suno and Udio
TechCrunch ↗
04
Reach signs Google AI deal, warns OpenAI.
Digiday ↗
05
USA Today Co sues OpenAI seeking $250m
Press Gazette ↗
06
Appeals court: training AI on 2,243 headnotes was not fair use
U.S. Court of Appeals for the Third Circuit ↗
The panel on this signal
They agreed on
–Settlements are pricing training data for the first time, and licensing rails are forming as the direct hedge.
–Provenance of training data now belongs to legal diligence.
–Music settled first, and its per-use model is the template others watch.
The competition economist disagreed on
–A negotiated settlement reflects the bargaining leverage of one litigant against one defendant; extrapolating a generalisable market-clearing price for training-data licensing from a single case overstates what one settlement can establish.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.