Dozens of lawyers have cited nonexistent case law — invented by AI — in court filings. A California court fined some of them, and several federal judges personally caught factual errors in AI-drafted documents. In the same period, legal AI startup Harvey raised $200 million at an $11 billion valuation in March 2026, and a month later Legora closed a $600 million Series D that made it a $5.6 billion company.

The money kept flowing while the accidents piled up. The fact that these two currents are running at the same time explains what is happening in the legal market right now.

Case Research, Briefs, Document Review — Where Law Firms Spend Their Hours Is Changing

In May 2026, Anthropic unveiled an AI tool built specifically for legal practices. Until now, the legal AI market belonged to specialized startups like Harvey and Legora, which got there first. Anthropic's direct entry signals that the market is moving past the startup-experiment phase and into platform competition. Anthropic began as an AI safety research organization and operates Claude, a general-purpose AI model. That organization has now started shipping tools aimed at a single industry.

The tasks it targets for automation are specific: document search and review, case-law research, deposition prep, and document drafting. These are the jobs that consume much of a junior associate's or legal clerk's day. Law firms have always had positions where this work tapers off as careers advance — and if AI compresses those hours, the scope of those positions changes with them.

The tool isn't standalone. It works in concert with Docusign, Box, and Thomson Reuters Westlaw, handling contract-management platforms, file repositories, and legal databases in a single workflow. Its coverage spans commercial law, privacy, corporate law, employment law, product-related law, and AI governance.

Harvey and Legora already have contracts with major firms and are in production. An Anthropic spokesperson said the legal sector "faces pressure to adopt AI, and the firms that move are pulling ahead fast." Firms that follow later won't just pay the adoption cost — they'll also have to close the speed gap with competing firms.

The AI Made Up Case Law — and the Market Grew Anyway

Skepticism about legal AI is not an unfounded worry.

The incidents reported in U.S. courts follow a similar pattern: a lawyer filed an AI-generated brief without reviewing it thoroughly, and the brief cited case law that does not exist. It came out in court, and a California court imposed specific fines on some of those lawyers. Several federal judges have acknowledged personally finding factual errors in AI-drafted documents.

The critics' point is concrete. An error in a legal document is not like a typo in a contract or an awkward phrase in an email. One bad case citation can change the outcome of a lawsuit and, in some cases, trigger disciplinary proceedings. Law is among the fields with the narrowest tolerance for error. A tool whose accuracy has not been fully validated is moving into that field fast, and some worry that the race for speed is outrunning quality control. Indeed, the tendency of AI language models to present nonexistent information as if it were real is, in law, not a mere technical limitation but a professional hazard.

Yet there is a reason the investment keeps coming. If a competing firm spends two hours on document review while your team spends twenty, that gap shows up directly on the fee estimate. In fiercely competitive markets, speed gets chosen over safety again and again. Delaying AI adoption has started to carry a cost of its own.

Research, Drafts, Document Review — You Don't Have to Be a Lawyer

What is happening at law firms is playing out the same way across other kinds of knowledge work.

Translate the tasks AI targeted first in law — document search and review, case research, drafting — into another vocabulary, and you get the daily work of a solo consultant, an independent product manager, or a corporate planner: reviewing materials, hunting down references, drafting proposals. Only the industry differs; the shape of the work overlaps.

There is a way to audit your own work right now. Measure how much of your day goes to gathering and organizing information. The more time you spend searching, compiling, and summarizing, the sooner AI adoption pays off. The same goes for how long it takes to produce the first version of a proposal or a plan. If a first draft takes three hours, the starting point is to find out whether AI can cut that to thirty minutes.

But you have to price in the cost of errors, too. A field where one line of a contract can lead to a lawsuit and a field where a proposal draft contains an awkward phrase have very different tolerances. The right depth of AI adoption depends on what an error costs in each task.

People who have spent years in established fields can review AI output against standards they already hold. The lawyers fined in California skipped that review. Law is giving the rest of us an early look at how outcomes diverge between those who keep their judgment while taking the speed, and those who set judgment down to get it.

While Harvey was being valued at $11 billion, a California court was fining lawyers who filed AI-written briefs without checking them. Both happened in the same year. From the moment Anthropic entered legal services, the price of giving up either speed or judgment started becoming visible — in hard numbers — across document-centric knowledge work everywhere.