Blueprint #37: The Andersen Jury Timeline Mess

Reid MercerSo I'm reading this filing at, like, midnight, and I'm thinking, "Grant, we called this."

GrantWe did call this. Twelve people, a jury box, actual humans deciding if a model is a copy machine.

Reid MercerThis is Blueprint. I'm Reid Mercer, that's Grant, and we've been circling Andersen versus Stability AI since episode thirty-six.

GrantFeels like forever ago.

Reid MercerRight? We previewed it, we said, "Watch this case," and now, plot twist, the jury's supposedly seated.

GrantSupposedly.

Reid MercerHere we go.

GrantI'm just saying, every headline this week says, "Trial begins," and none of them agree what that actually means today versus what happens next week.

Reid MercerOkay, but the opening statements thing is real.

GrantProbably real.

Reid MercerWow, ringing endorsement.

GrantI've been burned before, buying a car because a ad said, "One owner," and finding three names on the title.

Reid MercerSure, tie it back to cars.

GrantI'm saying, check the paperwork before you get excited.

Reid MercerFine. So before we build a whole episode on the jury is in, maybe we actually pull the filing and see if that headline holds up.

GrantNow we're talking.

Reid MercerOkay, so back up for a second, because if you skipped Blueprint thirty-six, you need the theory before you need the headline.

GrantRight, the actual legal question.

Reid MercerJudge Orrick, back in twenty twenty-four, ruling on the motion to dismiss, basically opened the door. Law firm analysis of that ruling says he found it plausible that copying images to train an AI system may be infringement.

GrantPlausible.

Reid MercerNot proven. Exactly. Plausible enough to survive dismissal and get to a jury. That's the whole doctrine everyone's calling model as copy.

GrantSo it's not that Orrick decided training is illegal, he just decided the argument gets a trial instead of getting thrown out.

Reid MercerRight. That's all a motion to dismiss ruling really does. It's not a verdict, it's a green light.

GrantAnd now twelve regular people get to decide if that's actually true.

Reid MercerWhich, dude, is wild when you remember how this case even started.

GrantSarah Andersen.

Reid MercerThe cartoonist. Back in twenty twenty-two, she wrote a New York Times essay. This thing went everywhere, and she said when she typed her own name into Stable Diffusion, it spit out images with, quote, "signature elements of her comics."

GrantHer own style coming back at her.

Reid MercerThat's the essay that basically lit the fuse for the whole lawsuit.

GrantAnd that's just one artist's search. Multiply that by every visual artist who tried typing their own name into the same box.

Reid MercerThat's the part that made it viral. It wasn't a legal brief, it was people watching their own style come back at them from a machine.

GrantSo we're not talking abstract training data philosophy anymore.

Reid MercerNo.

GrantSo we're talking about an artist who typed her own name and watched her own line work come out the other side of a machine.

Reid MercerThat's the image everyone remembers from this case.

GrantAnd for anyone tuning in fresh, this stopped being a law professor debate the second a jury got asked to weigh in.

Reid MercerRight, that's the shift.

GrantBlueprint Thirty-Six was us and a bunch of filings. This is Twelve people in a Box deciding whether Orrick's theory holds up in the real world.

Reid MercerWhich is a completely different animal.

GrantCompletely.

Reid MercerLaw professors argue in footnotes, a jury just says yes or no.

GrantAnd that yes or no becomes the number everyone in the industry starts quoting.

Reid MercerOkay, so that's the theory, that's the origin story. Now, here's the part I actually wanna get into. What actually happened in the courtroom this week? Because the headlines are moving fast, and I'm not sure they all agree with each other.

GrantUh-oh.

Reid MercerYeah, wait for it. Okay, so per the actual reporting, not last week's promo copy, the trial in San Francisco moved from discovery fights straight into opening statements this week.

GrantThat's the part I wanna see in writing.

Reid MercerIt's writing enough. Sigma Law Group's writeup on this pegs it as the first U.S. jury verdict anybody's gonna get on the model as copy theory.

GrantFirst one, not a verdict, the verdict.

Reid MercerRight, because there's also a Lanham Act claim riding shotgun on this, a trademark angle a fair use defense can't even touch.

GrantWhat's the trademark angle, though? Is that about her name generating knockoff art?

Reid MercerThe writeup doesn't spell it out beyond calling it a Lanham Act claim, but yeah, it's tied up with using her identity, not just the images themselves.

GrantHmm.

Reid MercerAnd Grant, think about the room. This is a theory that's lived in law review footnotes and one judge's opinion for two years. Now it's Twelve people instead of law professors deciding whether it's real.

GrantNo pressure.

Reid MercerNone at all.

GrantOkay, walk me through it. Opening Statements happened when exactly?

Reid MercerThe writeup says it goes to a jury September Eighth.

GrantSeptember Eighth.

Reid MercerThat's the date in the piece.

GrantYou know me, I don't move on anything without seeing paperwork first. That's true for cars, that's true for lawsuits. Hold On.

Reid MercerWhat?

GrantWe just said Opening Statements happened this Week, present tense, like it's already underway.

Reid MercerRight, that's the headline.

GrantBut the same writeup you just read me says it goes to a jury on the eighth. That's not already started, that's a scheduled date.

Reid MercerI mean, the eighth was this week.

GrantSure, but goes to and kicked off and reached opening statements aren't the same sentence. One's a calendar entry, one's a play-by-play.

Reid MercerHuh. Yeah, actually, nobody in what we've got describes an actual opening statement happening, just that date.

GrantThat's not nothing. If we're about to spend twenty minutes talking about what a jury supposedly heard, I wanna know they were actually in a room hearing it.

Reid MercerFair.

GrantIt's not me being difficult. If we say trial started, and it turns out it's trial scheduled, that's a fact we can't take back once it's out there.

Reid MercerI don't love that, but you're right.

GrantSo before we build the back half of this episode on trials underway, I wanna go back and check what the actual docket says versus what got typed into a blog post on the sixth. So I went back and actually read the case management filing instead of just the news write-up.

Speaker 3And?

GrantJanuary 13th of this year, both sides signed off on a joint case management statement.

Speaker 3Okay.

GrantIt says the case is set for April 5th, twenty twenty-seven.

Speaker 3Wait, next year?

GrantNext year, and plaintiffs are telling the court to plan for a fourteen-day trial.

Speaker 3Fourteen days? That's not a quick in and out. That's a whole production.

GrantWhich is exactly why trial open this week doesn't sit right with me.

Speaker 3Could be amended, though. Scheduling orders get pushed constantly, especially in a case this size.

GrantTrue, but amended isn't nothing. That means a stipulation gets filed, or one side moves for it and a judge signs an order. There's paperwork either way.

Speaker 3Right, and paperwork like that shows up on the docket. It's not invisible.

GrantSure, but pushed forward, not backward a full year. That's the part I can't explain.

Speaker 3So what do we actually know for certain?

GrantI pulled the docket itself. Official record, not a write-up, not a press release.

Speaker 3And what's on it?

GrantLast entry we can confirm is from September first. No trial transcript, no jury verdict sheet, nothing that screams opening statements happened.

Speaker 3So we've got a filing that says next spring and a docket that just kind of stops.

GrantRight. Neither one confirms a jury sat down this week.

Speaker 3I still think you're overreading the gap. Big federal cases get their dates moved all the time. A judge clears their calendar, a settlement talk falls through, and suddenly April becomes September.

GrantMaybe, but I'm not going on air acting like the trial definitely started when the paperwork I can actually check doesn't back that up.

Speaker 3Fair. So we say it plain. We don't know for certain which version is true. That's the honest answer, and it's a better answer than pretending we cracked it.

GrantIt's also just good practice. You say, "We checked, and here's what we found," instead of repeating a press release like it's gospel. Which, by the way, is the whole reason I don't trust a single trial starts headline without pulling the docket myself. Blog posts get the date right and the verb tense wrong all the time.

Speaker 3Noted for the record, Grant's not buying the tidy headline.

GrantSomebody's got to check the paperwork.

Speaker 3All right, but here's the thing that actually matters more than the exact date on the gavel.

GrantGo on.

Speaker 3Whether it's this month or next spring, the theory being tested doesn't change. Judges already said the copying claim can go forward.

GrantRight. The calendar's a mess. The stakes underneath it aren't.

Speaker 3And that's the frame I want listeners to hold on to. Whatever the actual court date turns out to be, or explaining on the theory itself doesn't get unwound by a scheduling mix-up.

GrantRight. The calendar's confusing, but the legal question the jury eventually answers is exactly the one we've been describing.

Speaker 3Exactly. So let's talk about what happens the day twelve people actually rule on it.

GrantBecause a plaintiff win doesn't stay contained to one case.

Speaker 3Not even close. It becomes the reference point for every image model lawsuit sitting behind it. Okay, so say the jury actually buys the model as copy theory. Plaintiffs win.

GrantThen every case sitting behind this one just got a lot more dangerous for the defense side.

Speaker 3Every Midjourney case, every Runway case, even the AI vendors the studios themselves quietly signed up.

GrantWait, that's the part people keep skipping. This isn't just a Stability AI problem anymore.

Speaker 3Right, because the whole legal theory here is what training a model on someone's work actually is under copyright law.

GrantAnd if a jury signs off on training equals copying, that reasoning doesn't stay in this courtroom.

Speaker 3It walks straight into every other pending fight built on the same argument. There's a guy, Vincent Couey, who keeps a running list of basically every live creative AI copyright fight, images, music, all of it.

GrantOh, I've seen that thing. It's huge.

Speaker 3It's a spreadsheet worth of lawsuits, and this verdict becomes the template a lot of those other cases get measured against.

GrantWhat's actually on that tracker besides case names? Does it show which ones are furthest along?

Speaker 3Status and stakes, yeah. Some are barely past a complaint. Some are deep into discovery. This one's just further along than most.

GrantSo if you're Midjourney's lawyers right now, you're watching this trial like it's your own.

Speaker 3Hundred percent, because a plaintiff win here doesn't just cost Stability AI money.

GrantIt resets what fair use even means for training data across the entire industry.

Speaker 3Sure, eventually, but you're skipping the part where every smaller AI shop can't afford that fight.

GrantThey can't afford litigation, no.

Speaker 3So it's not the big guys who get hurt first. Stability, Midjourney, they'll survive appeals for years.

GrantIt's the smaller vendors who get frozen out while the case law sorts itself out.

Speaker 3Give me an example of what that actually looks like for a smaller shop.

GrantPicture a two-person startup that fine-tuned an open model for a niche use case. They don't have the legal budget to defend a downstream taint theory, so they just shut down or get acquired for parts.

Speaker 3Exactly. You get a licensing regime built for giants and a graveyard of tools that never made it to market.

GrantI'd take that trade over the current setup where nobody's getting paid at all.

Speaker 3Of course you would. You're the optimist in this friendship.

GrantSomebody's got to be.

Speaker 3Fair, but either way, we're talking appeals stretching years past whatever verdict comes down next.

GrantYears of uncertainty for anyone building an image model right now.

Speaker 3Which, funny enough, is exactly the position Hollywood studios are sitting in too.

GrantRight. Does any of this actually touch the deals the studios already signed?

Speaker 4So does any of this touch the deals studios have already signed? The ByteDance thing we broke down on Blueprint thirty-four, that's a handshake, not a license.

Speaker 3It doesn't matter whose handshake it is. If the jury buys model as copy, the exposure runs to anyone using the tainted model, not just Stability.

Speaker 4Wait, so the MOU is just paper?

Speaker 3Worse, it's a memorandum of understanding. No indemnification language, no warranty the training data is clean, just two companies agreeing to agree later.

Speaker 4So what would actual protection even look like on paper?

Speaker 3A real license would need reps and warranties that the training data is clean, an indemnification clause that puts the financial risk on the AI vendor, not the studio, and probably an audit right so the studio can actually check.

GrantAnd the MOU has none of that?

Reid MercerNone of it. It's a press release document dressed up as a legal one.

GrantUgh, of course.

Reid MercerAnd remember what the actual complaint alleges? Stable Diffusion and the tools built on top of it were trained on billions of scraped images, plaintiff's work included, and the output can reflect it.

GrantBillions with a B.

Reid MercerBillions. That's not a rounding error, that's the whole training set.

GrantOkay, but ByteDance isn't Stability. Different company, different model.

Reid MercerDoesn't matter if it's downstream. That's the phrase doing all the work. Anything built on a tainted base inherits the taint.

GrantDoes downstream have a legal definition here, or is that just the phrase you and I are using?

Reid MercerIt's shorthand. The actual argument would be that any tool built on the tainted weights is itself infringing because it's using a copy of a copy functionally. Courts haven't drawn a bright line on how many steps removed still counts.

GrantSo a studio three deals removed from Stability could still be in the blast radius?

Reid MercerCould be. Nobody's tested exactly how far it stretches yet. That's part of what makes this verdict so consequential.

GrantSo every studio that's signed one of these quiet AI partnerships-

Reid MercerIs one ruling away from finding out their partner's foundation model has the same problem Stability does.

GrantHuh, that's a scarier headline than the trial itself.

Reid MercerIt's the one nobody's writing yet.

GrantAh, great. So the MOU doesn't buy them anything.

Reid MercerNot a force field, not even an umbrella.

GrantFine, fine. No umbrella.

Reid MercerNow, here's the part that gets messier for anyone hoping U.S. law settles this cleanly.

GrantThere's more?

Reid MercerThere's a court across the Atlantic that already ruled on a version of this exact question.

GrantSame theory, different country?

Reid MercerSame theory, different answer, and it's not the one Silicon Valley wants.

GrantOkay, now I need to hear this.

Reid MercerSo while we've been chasing dockets in San Francisco, London already ruled on this exact fight.

GrantWait, same theory, different country?

Reid MercerSame theory. Getty Images sued Stability over the same model, same training data, and the UK High Court ruled back on November fourth of last year.

GrantAnd?

Reid MercerThe court said the weights themselves aren't an infringing copy under UK law, full stop.

GrantShut up.

Reid MercerYeah. Latham & Watkins wrote it up. Getty's secondary infringement claim got rejected because the weights aren't treated as a copy of the training images.

GrantSo the exact argument that got Andersen passed a motion to dismiss here just got tossed on site over there?

Reid MercerPretty much. Two courts, same underlying tech, opposite starting points on what a model Even IS legally.

GrantThat's not a small gap. That's the whole architecture of the case splitting depending on which side of the Atlantic you're standing on.

Reid MercerPart of it's just structural. The UK claim was framed as secondary infringement, which asks whether an infringing copy was imported or dealt with in the country. The U.S. theory is framed around the reproduction right itself, whether the training process makes a copy in the first place.

GrantSo they're not even arguing the exact legal question.

Reid MercerRelated question, different doorway into it, which is part of why a U.S. jury siding with plaintiffs doesn't automatically mean the UK court got it wrong. They were answering something adjacent.

GrantAdjacent, but pointed in opposite directions on the same underlying tech.

Reid MercerAnd Stability is not some regional player. It's the same company, same weights, getting judged two different ways, which means if Andersen's jury sides with the plaintiffs, you've got a model that's legal to ship in the UK and radioactive to touch in the U.S. for the exact same file.

GrantRight, because every vendor selling into Hollywood is selling globally. You can't build one version of Stable Diffusion for California and a different one for London.

Reid MercerUnless you start regionally gating model access, which nobody's built for and nobody wants to pay for. And I'll say it plainly, I don't think we call the Andersen verdict from this UK ruling. Different statute, different court, different judge entirely.

GrantAgreed. It tells you the doctrine is unsettled, not which way it breaks here.

Reid MercerSo here's what we're actually watching before the next episode.

GrantThe docket, for one. Did an opening statement happen? And when is the next confirmed filing?

Reid MercerAnd whether Andersen jury instructions even reference model weights the way the UK opinion did, or whether the U.S. Theory stays purely about the training copy.

GrantAnd if any studio quietly renegotiates AI vendor language after this UK decision gets more attention.

Reid MercerThat's the checklist: docket, jury instructions, contract language.

GrantThree things, one verdict pending.

Reid MercerWe'll have updates the second any of it moves. So where does that leave us? Docket says one thing, filing says another, and the actual jury, still a question mark.

GrantYeah, we're not resolving that today.

Reid MercerNope. We'll keep an eye on it till the paperwork catches up with the headlines.

GrantOr the headlines catch it with the paperwork.

Reid MercerFair.

GrantEither way, next time we've got an update, you'll hear it first.

Reid MercerPromise.

GrantIf you got thoughts on any of this, the jury, the UK ruling, whatever, email us, Blueprint@hemato.com.

Reid MercerTag us on social too. We read all of it.

GrantSend it to a coworker who still thinks AI lawsuits are boring.

Reid MercerTheyre not boring. Theyre just slow.

GrantNew episodes every Tuesday.

Reid MercerWell be back when the docket actually says something.

GrantTake care, everybody.