A Second Nine-Figure AI Training-Data Copyright Settlement Will Land by the End of 2026

The Prediction

By December 31, 2026, at least one AI company other than Anthropic will publicly agree to a copyright settlement or class resolution of $100 million or more (a nine-figure sum) tied to the use of copyrighted works as training data. The settlement must be reported by a credible outlet or court filing, must be distinct from the ~$1.5 billion Bartz v. Anthropic settlement, and must be characterized primarily as compensation for training-data copyright claims (not a general commercial dispute or a routine content-licensing deal struck outside litigation).

Confidence: 58% · Tier 2 (mid-term) · Target: December 31, 2026

Why I Believe This

The conditions that produced the first nine-figure settlement have not gone away — they have intensified. Bartz established that a frontier lab will write a very large check rather than take a piracy claim to a jury, and it gave every plaintiff's lawyer in the country a public benchmark of roughly $3,000 per work to anchor a demand. That benchmark is the mechanism I described in the price of training data and AI's copyright reckoning: once a number exists, every other case gets measured against it.

The pipeline is unusually full. More than 70 AI copyright suits are active in U.S. courts as of mid-2026, spanning authors, news publishers, visual artists, and music labels, with aggregate claimed exposure north of $50 billion. Several involve defendants with the balance sheets to settle at nine figures and the incentive to make litigation risk disappear before going to public markets or closing a funding round. The marquee case — The New York Times v. OpenAI and Microsoft — is exactly the kind of matter that resolves in a large settlement when the discovery and trial risk becomes unattractive to both sides.

There is also a strong structural incentive to settle in 2026 specifically. With the first appellate fair-use ruling pending from the Third Circuit, defendants face genuine uncertainty about whether their core legal shield survives review. A known settlement number is preferable to an unknown appellate rule, and the rational move for a well-capitalized lab is to convert that uncertainty into a fixed cost before the doctrine hardens against it.

What Would Falsify This

This prediction fails if, by the end of 2026:

• No AI company other than Anthropic agrees to a training-data copyright settlement of $100 million or more; or
• The only nine-figure resolutions are ordinary licensing deals struck outside active litigation (those are real and growing, but they are not what this predicts); or
• Major cases instead proceed to trial, get dismissed, or settle only at sub-nine-figure amounts.

The most likely way I am wrong is timing: the largest cases could grind into 2027 before resolving, or defendants could choose to litigate the fair-use question to final judgment now that an appellate ruling is imminent. A favorable Third Circuit decision for AI developers would also reduce settlement pressure across the board and could push the next big settlement past the deadline.

How I'll Evaluate

On or shortly after December 31, 2026, I will check whether a qualifying settlement was publicly reported, confirm the dollar threshold and that it is distinct from Bartz, and confirm it is characterized as a training-data copyright resolution. Partial credit reasoning will be noted if a settlement is agreed in principle but the amount is unconfirmed at year-end.

Related Analysis

• The price of training data: AI's copyright reckoning arrives
• The UK kills its AI copyright opt-out and reopens the licensing battle