Mark Lanier took on Meta and Google in a landmark California case, facing almost unlimited resources, enormous legal teams and some of the most advanced technology in the world. Yet the US trial lawyer defeated both companies – with the help of AI.
Lanier represented a young woman who said she had used YouTube and Instagram since childhood and had later developed anxiety, depression and body-image disorders. The case was not principally about specific harmful content. Instead, the plaintiff argued that the platforms themselves had deliberately been designed to make children and teenagers dependent on them.
On 25 March 2026, the jury awarded the woman a total of $6m – $3m in compensatory damages and a further $3m in punitive damages. Meta and Google were found liable on every count.
One David, Two Goliaths
The Guardian subsequently published a portrait of Lanier, casting him as a modern-day David who had taken on two Goliaths of the technology industry and won. Most revealing was his account of how he had used artificial intelligence during the trial.
The real divide is not between those who use AI and those who do not, but between those who let it think for them and those who know how to make it serve their own judgment.
Using AI effectively first requires a person who understands the system, has thought carefully about its purpose and knows how to direct it. For the trial, Lanier used a system developed specifically for him by BoodleBox. It combined several large language models, including ChatGPT, Claude and Gemini.
How Lanier Put AI to Work
That combination was an important part of the strategy. Effective use of AI should draw on several models wherever possible. Any individual system can hallucinate or produce flawed results because of gaps in its training data.
Lanier said AI had allowed him to complete in roughly 10 hours what would previously have taken 30. He compared the technology to additional, well-trained employees who knew the case files and could work around the clock.
Lanier, who is a Christian, said AI could be used for good or evil – just like the legal system itself.
He identified around 30 tasks that AI could perform during the case, but four areas proved particularly important: document analysis, the evaluation of daily court transcripts, preparation for cross-examination and the simulation of jury reactions.
The first was document analysis. A case of this scale generates an immense volume of material, including internal company records, emails, product documentation, witness statements, expert reports and legal submissions.
The AI served as an assistant, locating documents relevant to particular arguments and identifying connections across the evidence. Rather than replacing the lawyers’ judgment, it gave them faster and more systematic access to the material on which that judgment depended.
Preparing Overnight
The second task was analyzing the daily court transcripts. After each day of testimony, Lanier’s team could feed the transcripts, witness statements and other new developments into the models. Overnight, the system helped them identify arguments and possible lines of response for the following day.
The trial lasted several weeks, forcing the plaintiff’s lawyers to respond quickly to witness testimony and arguments from the defense. The models helped them assess what had happened in court that day and what it might mean for the next stage of the case.
The third area was preparing cross-examinations, a central feature of the adversarial Anglo-American court system.
Lanier particularly emphasized his preparations for questioning Mark Zuckerberg. The AI brought together Zuckerberg’s previous statements, internal documents and potential contradictions, giving Lanier a concentrated body of material to work with on the morning of the questioning.
Building a Virtual Jury
The fourth and most striking use of AI was to simulate the jury’s reactions.
Lanier used language models to test how jurors might respond to arguments and witness testimony. The AI did not know the actual jurors, of course, and any improper access to their identities or private lives would have been prohibited.
Yet analyzing a jury as closely as the rules permit has always been part of an American trial lawyer’s work. The aim is to understand which arguments may resonate, which may seem too technical and which documents are most likely to persuade.
Such strategic testing would traditionally have relied on discussions among lawyers. AI allowed Lanier to take it much further.
In The Guardian’s portrait, he explained how much information the legal teams could gather about prospective jurors during the selection process, despite their anonymity outside the courtroom.
“We have questionnaires they filled out that tell us their age, their gender, their occupational history, their family status.” This provided the team with an initial picture. The questionnaires also included questions such as: “Who are three people you most admire and why? Who are three you least admire and why? How do you feel about this or that on a scale of one to 10?”
By the end of the selection process, the lawyers had assembled a detailed file on each juror.
Lanier’s AI models then used this information to simulate the individual jurors, allowing his team to test potential arguments against virtual versions of each one.
At the end of every day in court, Lanier said, the latest transcripts were fed to his AI shadow jury. The team could then ask: What did juror number 11 think of that witness? What did juror number 7 consider important? Where had juror number 3 become confused?
Lanier summed up the system in two words: “Pretty cool.”
Beating Big Tech at Its Own Game
The imbalance between the two sides shows how powerful a well-designed AI system can be.
“The opposing side”, Lanier said, “had unlimited resources. They had dozens of lawyers in the courtroom.”
He knew that his opponents were prepared to use every advantage at their disposal.
That naturally included artificial intelligence. Google and Meta each have their own AI systems, including Gemini and Meta AI. They are also likely to have access internally to unreleased models, more advanced configurations and specialized tools unavailable to ordinary users.
Lanier was determined to turn that technological advantage against them. He succeeded.
A Big Tobacco Moment
The Guardian described the verdict as a big tobacco moment for big tech. As with tobacco and pharmaceutical products, the central question is whether major companies knowingly designed their social media platforms to foster dependence and cause harm.
The jury’s decision was emphatic. On 25 March 2026, it awarded the plaintiff $6m, split evenly between compensatory and punitive damages, and found both Meta and Google liable on every count.
According to press reports, more than 2,000 similar lawsuits against social media companies are still pending in the United States.
The world’s most powerful technology companies entered the courtroom with vast legal teams, near-limitless resources and artificial intelligence of their own. Lanier’s victory showed that, in the right hands, the same technology can narrow even the widest imbalance of power.