What AI did Mark Lanier use in the Meta social media addiction trial?

Lanier's team used a multi-model platform that put several leading models in one workspace, on a custom license reported to cost six figures a year. They fed each day's transcripts to the models for evaluation, used AI to sharpen how arguments were phrased, and ran the jury's written questions through models during deliberations to read where the panel stood. He set firm limits: no AI-written briefs and no unsupervised legal research.

Do you need a six figure budget to use AI like the top trial firms?

No. The six figure number was the cost of a custom cloud license. The same core capability, transcript analysis, document search, and persuasive drafting support, can be built in house on a single GPU machine for a fraction of that, and it keeps your data private. The point is to start small with one tool that saves real hours, then add from there.

A Lawyer's Guide to AI: What Mark Lanier Did to Beat Meta, and How to Build It In House

Q: Can a law firm use AI without risking client confidentiality?

Yes, but how you deploy it decides the answer. Public, consumer AI tools send your prompts to a vendor's servers, which is a real problem when those prompts contain privileged facts and client data. The safe path is an in-house system that runs on hardware your firm controls, so case material is processed inside the building and never leaves it. That is the model we build for plaintiff firms.

Q: Is it safe to put privileged case files into ChatGPT or other public AI tools?

Treat it as risky. Pasting privileged work product, client identifiers, or medical records into a public tool means that material leaves your control. For anything privileged, use a system your firm hosts and controls, where nothing is sent to an outside server. Save public tools for general, non-confidential work.

A trial lawyer just used AI to win a landmark verdict against Meta. The lesson for the rest of us is not that AI is magic. It is that AI for law firms is leverage you can own, if you build it the right way and keep your data inside the building.

By Jacob Malherbe · June 25, 2026

I wrote A Lawyer's Guide to AI because most of the conversation in our space is stuck in two unhelpful camps. One camp is scared of AI and treats it like a liability waiting to happen. The other hands it the wheel and ends up in the headlines for the wrong reasons. The truth sits in the middle, and the clearest proof of it came out of a Los Angeles courtroom. Here is what happened, why the way you use AI for law firms matters more than whether you use it, and how to build your own version in house so your privileged work never leaves the building.

The verdict every plaintiff firm should study

In the social media addiction trial, veteran trial lawyer Mark Lanier won a $6 million verdict against Meta and Google, three million in compensatory damages and three million in punitive damages, with the jury finding the platforms negligent and dangerous. Meta was held seventy percent responsible and YouTube thirty percent. It is a bellwether for the broader social media litigation, and the firms paying attention to how he did it are the ones who will be ready for what comes next.

Lanier was open about the role AI played. He described it as like having "10 additional workers" who know the file inside and out and work around the clock, and said it let his team do roughly thirty hours of work in ten. That is the part most people repeat. The part most people skip is the discipline behind it, and that discipline is the whole lesson.

AI is leverage, not the lawyer

Lanier did not use AI to write his briefs or to run unsupervised legal research. Those are the exact uses that have produced a wave of sanctions and fabricated citations across the country. He used it as horsepower under his own judgment. He even caught the AI making one incorrect claim from the record and corrected it, and his line about the whole arrangement was simple: "It is not unbridled." You are still the important part of the equation.

That is the message of the book in one sentence. Do not be scared of AI, and do not worship it either. Learn to use it. It multiplies forty plus years of trial instinct. It does not manufacture instinct you do not have. For a plaintiff firm, AI for law firms should mean a faster, sharper version of the lawyers you already are, not a replacement for the judgment that wins cases.

The part nobody warns you about: where your data goes

Here is where I want to push the conversation past the highlight reel. The tool Lanier's team used was a cloud platform, licensed for a six figure annual fee, that routed their work through several models hosted by outside vendors. For a trial team with that budget and a compliance wrapper around it, fine. For the typical small or mid size plaintiff firm reaching for the nearest public tool, it is a real problem.

When you paste a privileged memo, a client's medical records, or your trial strategy into a public AI tool, that material leaves your office and lands on someone else's servers. You no longer control where it sits, who can see it, or how it gets used to train the next model. For a profession built on confidentiality and privilege, that should stop you cold. The convenience is not worth handing your client's file to a third party. This is the single biggest mistake I see firms making with AI right now, and it is completely avoidable.

The in house answer: nothing leaves the building

The fix is not to avoid AI. The fix is to bring it in house. When the model runs on hardware your firm owns and controls, your case material is processed inside the building and never goes out to a public service. You get the leverage Lanier described without the exposure, and you stop paying a six figure subscription for the privilege of sending your data away. This is the core idea behind everything we build at MTAA, and it is the natural extension of the book. Own the tool, own the data, keep the privilege intact.

What Lanier's team actually did, step by step

Strip away the brand names and the playbook is straightforward. You can recreate all of it.

Nightly transcript review. At the end of each court day, the team fed that day's transcripts to the models and asked for an evaluation: what landed, what did not, where a witness opened a door.
Sharper phrasing. They used AI to find more persuasive, more visceral ways to make a point, then a human picked the version that fit the case and the room.
Reading the jury. During deliberations, the jury's written questions went into the models to help gauge where the panel's head was, so the team could anticipate rather than guess.
An overnight war room. Tasks were handed off at night and finished by morning, so the lead lawyer walked in each day with the work already done and ready to review.

None of that requires a six figure license. It requires a system that can read your documents, answer questions against them, and draft on demand, all without sending a word outside your office.

How to build your own in house version

This is the part I refuse to gatekeep, because the firms that build it will be far better off than the ones who wait. Here is the blueprint for an in house system that does what Lanier's did, on hardware you control.

One GPU machine on your network. A single capable box, sitting in your office, is enough to run a strong open model locally. This is the foundation. Nothing it processes leaves the building.
An open model you host yourself. Modern open models are good enough for transcript analysis, document question and answer, and drafting support. Because you host it, there is no per seat cloud fee and no data leaving your control.
Ingest your documents and transcripts. Load pleadings, depositions, daily transcripts, and case files into a private knowledge layer so you can ask plain English questions and get answers grounded in your own record.
Make your video searchable. Run trial footage and deposition video through local transcription so every minute becomes searchable by timestamp. Ask for the moment a witness said a specific thing and jump straight to it.
A one way research fetch. Let the system pull public information inward, news, MDL updates, public filings, while a hard rule keeps privileged case facts from ever going outbound. Research comes in. Your client's file never goes out.
Start with one small tool. Do not try to build all of this in a week. Build the single tool that saves you the most hours first, maybe transcript question and answer, get the firm using it, then add the next piece.

That is a real, private version of what won against Meta, and it is within reach of a normal plaintiff firm. If your team does not want to assemble it from scratch, this is exactly the kind of in house system we build and stand up for firms, but I would rather you build a small one yourself than keep pasting privileged files into a public box.

The same system finds your next clients

Here is the part that ties AI for law firms back to growth. The same in house setup that helps you try a case also helps you fill the top of your funnel. The tools we build for client acquisition, the scanners that surface emerging torts, the intake assistants that qualify claimants around the clock, the systems that watch the litigation landscape for new opportunities, run on the same foundation. Once you have a private model and a way to feed it your own data, you can point it at intake and marketing as easily as at trial prep. The firm that owns its AI owns both halves: better cases and more of them.

Start small, and do not be scared

The lesson of the Meta verdict is not that you need a six figure subscription or a forty year career to use AI well. It is that AI for law firms works when a capable lawyer keeps both hands on the wheel, and when the tool is built so your client's trust is never the price of admission. Do not fear it. Learn it. Keep it in house. Build something small this quarter, own it completely, and let it compound. That is the whole guide, and the firms that take it seriously now will be the ones the rest are studying a year from now.

Frequently asked questions

Can a law firm use AI without risking client confidentiality?

Yes, but how you deploy it decides the answer. Public consumer tools send your prompts to a vendor's servers, which is a real problem when those prompts contain privileged facts. An in house system that runs on hardware your firm controls processes case material inside the building and never sends it out. That is the model we build for plaintiff firms.

What AI did Mark Lanier use in the Meta trial?

His team used a multi-model platform on a custom license reported at six figures a year. They fed each day's transcripts to the models, used AI to sharpen phrasing, and ran the jury's written questions through models during deliberations. He set firm limits: no AI-written briefs and no unsupervised legal research.

Do you need a six figure budget to use AI like the top firms?

No. That number was a cloud license. The same core capability can be built in house on a single GPU machine for a fraction of the cost, and it keeps your data private. Start with one tool that saves real hours, then add from there.

Is it safe to put privileged case files into public AI tools?

Treat it as risky. Pasting privileged work product or medical records into a public tool means that material leaves your control. For anything privileged, use a system your firm hosts and controls. Save public tools for general, non-confidential work.

Build your firm's AI the right way

If you want help standing up a private, in house AI system, or you want to talk through where to start, Mass Tort Ad Agency builds these for plaintiff firms and will give you a straight, no-pitch conversation about it. Talk to Jacob, or read more about AI for law firms and how to put it to work.

Trial details reflect public reporting on the social media addiction verdict as of 2026. This article is general information for law firm owners, not legal, ethical, or technical advice for a specific matter.