Training data is becoming a paid input and rights holders are booking real, high-margin revenue against archives that were previously dead weight — but the legal foundation is weaker than the narrative assumes, because the one substantive US fair-use ruling went the model developer's way on training itself.
📈 What changed: 21 Jul 2026: Judge Araceli Martinez-Olguin granted final approval to the Bartz v. Anthropic settlement, entering judgment and dismissing with prejudice — $1.5B across roughly 482,…
| Indicator | Score | Reading |
|---|---|---|
| Maturity higher is better · high confidence | 45/100 | First paid deployments are live and disclosed, but pricing is bespoke and no standard has formed. |
| Evidence Strength higher is better · high confidence | 75/100 | Audited segment disclosure at Wiley and SEC-filed corporate actions anchor the base; deal terms are often press-reported. |
| Commercial Proximity higher is better · high confidence | 66/100 | Real disclosed revenue at high margin, but small in absolute terms and lumpy between periods. |
| Capital & Policy Support higher is better · high confidence | 36/100 | No public funding; the policy lever is transparency and provenance obligation rather than money. |
| Crowding Risk lower is better · high confidence | 27/100 | Structurally unloved value names where the licensing option is not the reason most holders own them. |
| Reflexivity Risk lower is better · high confidence | 33/100 | Cash-generative underlying businesses, but the licensing line itself is announcement-driven and lumpy. |
Every indicator score is computed by the Northstar engine from analyst-set ordinal bands — never hand-written, never stored.
The public page is the summary (backed by 18 cited sources). Full briefing — company map, catalysts, native signals and sources — is inside the app.
Open in the appStart free →The Northstar Signal — what moved across all 350 themes and why, in your inbox. Free, no spam, one-click unsubscribe.