In partnership with
Hey readers,
A federal jury needed less than two hours to end one of the most closely watched AI lawsuits in history. Meanwhile, the case that could actually decide whether you get sued for last week's Midjourney render is just getting started — and it goes to trial in six weeks.
Six rulings landed across four courts in the last nine months. None of them agree with each other. Here's what actually changed, and what to do about it before the next one lands.
ADVERTISEMENT
The Counsel's Corner
The market intelligence most desks are missing.
CEOs, politicians, regulators, and analysts talk for hours on podcasts every day.
Radar actively transcribes 115,000+ podcasts and turns them into structured data, searchable within minutes of airing.
Track a company or person by name, with each mention cited to the exact quote, speaker, and timestamp, or search for commentary on a thesis or sector with semantic search.
Radar is built by former Twitter and Tesla engineers, using an AI-native transcription pipeline that delivers high accuracy and extensive data enrichment.
Back to the Briefing
AT A GLANCE
Four courts ruled four different ways on AI training in the last nine months. Getty lost, Anthropic paid $1.5B, and the trial that could decide whether individual artists can win starts September 8 — one week after the EU's own enforcement deadline hits.
THE STORY
Six Rulings, Zero Consensus
Start with the number everyone's citing: $1.5 billion. That's what Anthropic agreed to pay to settle Bartz v. Anthropic — the largest copyright payout in U.S. history, working out to roughly $3,000 per book across 7 million pirated titles used in training. It's not about images. But it proved something creators had been told not to expect: AI companies will write real checks when the paper trail is bad enough.
Two months earlier, on the other side of the Atlantic, Getty Images lost. The UK High Court rejected Getty's core copyright claim against Stability AI in November 2025, handing the company only a narrow trademark win on early Stable Diffusion outputs. Getty has permission to appeal. A separate US case is still in its opening rounds.
Meanwhile, a Delaware court became the first in the country to rule that training an AI on someone else's copyrighted material is not fair use — a decision now facing its first appellate review. And in Munich, a German court held OpenAI liable after ChatGPT reproduced song lyrics almost verbatim.
Four courts. Four different answers to the same basic question. That's the actual story here — not a bad ruling, but an unsettled one, with the case most likely to affect individual creators still months from a verdict.
THE SCORECARD
Where Every Case Actually Stands
$1.5B
Bartz v. Anthropic — largest AI copyright settlement ever
Sept 8
Andersen v. Stability AI, Midjourney & DeviantArt trial date
Aug 2
EU AI Act training-data enforcement takes effect
Case | Status | What it means for you |
|---|---|---|
Getty Images v. Stability AI | LOST · ON APPEAL | Training on scraped images isn't automatically infringement — yet |
Andersen v. Stability AI, Midjourney & DeviantArt | TRIAL SEPT 8 | First real test of whether individual artists, not just publishers, can win |
Disney/Universal/DreamWorks v. Midjourney | ONGOING | Generating recognizable characters is a sharper, separate risk from training data |
Thomson Reuters v. Ross Intelligence | ON APPEAL | Only ruling so far to reject fair use for AI training outright |
GEMA v. OpenAI | LIABLE | Output that reproduces protected work can be actionable even when training wasn't the issue |
WHAT CHANGED FOR YOU
The Fine Print Grew Teeth
Here's the part most creators skip past. In March 2026, the Supreme Court declined to hear Thaler v. Perlmutter, letting stand the rule that pure AI output can't be copyrighted at all — no exceptions, no workaround. That was already true. What's new is that platforms are getting sharper about who eats the risk if a generated image turns out to resemble something protected.
Midjourney's current terms hand you a commercial license if you're on a paid plan, require the Corporate tier once you clear $1 million in annual revenue, and explicitly do not indemnify you against infringement claims. If a client's lawyer comes calling, that's your problem, not Midjourney's. Adobe took the opposite bet: Firefly trains exclusively on licensed Adobe Stock content and public domain material, and Adobe indemnifies enterprise users against infringement claims tied to its output.
The gap isn't quality anymore. It's who's holding the bag if a court eventually rules the other way.
| Midjourney | Adobe Firefly |
|---|---|---|
Training data | Web-scraped (LAION and successors) | Licensed Adobe Stock + public domain |
Commercial license | Paid plans only | Paid plans only |
Infringement indemnification | None | Yes, for enterprise users |
Best fit | Concepting, personal work, social content | Client deliverables, brand and ad work |
THE PLAYBOOK
Three Moves Before September
01 Save your working files.
Prompt history, layers, and hand edits are your evidence of human authorship — the only thing that gives you any copyright claim at all post-Thaler.
02 Match the tool to the stakes.
Personal projects can absorb legacy-model risk. Anything a client will publish under their name deserves an indemnified tool, even at a higher price.
03 Watch September 8, not the headlines.
Settlements make news. The Andersen trial verdict is the first one that will actually tell individual creators, not just media companies, where they stand.
WHAT'S NEXT
My Read: Insurance Becomes the Product
Here's my prediction. Within the next 12 months, "indemnified" stops being fine print and becomes a line item brands actually shop for — the same way cloud buyers started demanding uptime SLAs once outages got expensive enough to matter. The AI image tools that can promise legal cover, not just better outputs, will win the enterprise budget. Everyone else ends up competing on price with creators who simply can't afford the risk.
Indemnification is about to become the actual product.
None of these six cases will feel finished by the time the next issue lands. That's kind of the point — this isn't a settled area of law you read once and file away, it's one you have to keep checking.
Reply and tell me: does legal cover factor into which AI tool you reach for on client work, or are you rolling the dice like the rest of us?
Talk soon,