In 2023, a New York lawyer named Steven Schwartz did something that would change the trajectory of the next three years of legal technology: he asked ChatGPT for case law to support a client’s brief, and it gave him six cases (court decisions) that didn’t exist. Despite this, he filed the brief anyway. Afterwards, a federal judge fined him $5,000 and ordered him to write a personal letter of apology to every judge he had falsely cited. This event became the industry’s cautionary tale, the story every lawyer now references when the conversation turns to AI.
Despite its issues, the cautionary tale turned into a pattern. A database maintained by a researcher at HEC Paris has now tracked over 1,200 court filings worldwide containing AI-fabricated citations or quotes (known as hallucinations). In May 2026, a federal judge in Oregon issued the largest penalty of its kind in U.S. legal history: $110,000 combined, against two lawyers who submitted 23 fake citations and 8 invented quotations in a single brief. Since then, more than 25 federal courts have required lawyers to formally certify that a human actually checked the AI-produced work.
Now that the uncomfortable half of the story is out of the way, the other half benefits legal work. Lawyer adoption of AI has grown fast, with adoption hitting 83% in June 2026, up from under 20% just three years earlier. AI contract review now cuts review time by up to 85% while also hitting roughly 95% accuracy, which actually beats the 80% accuracy of manual review. AI makes legal work faster and more accurate on average, yet it has triggered a massive wave of professional sanctions across the legal field. The sanctioned lawyers are using the same high-quality AI tools as everyone else, but they simply fail to verify the output before filing it in court.
The underlying shift is quieter than the sanctions headlines, and it’s the one that matters most for anyone picturing a legal career. For decades, the way one became a lawyer was by doing the unglamorous grunt work first– reviewing thousands of documents, drafting first-pass memos, running down citations. Though it wasn’t exciting or fulfilling, it built the necessary foundations. But that’s precisely the work AI now does way faster than a junior associate can. In fact, one legal-industry estimate suggests that if the technology performs as projected, 85-90% of the work that traditionally trained new lawyers could disappear.
Instead of ignoring that possibility, some firms are actually rebuilding around it. For example, Ropes & Gray, a global firm of roughly 1,500 lawyers, launched a program in late 2025 letting first-year associates put up to 20% of the annual billable-hour target (roughly 380 hours) toward learning AI, without the pressure of billing that time to a client. This is a firm admitting that the old training model is fading away, and building a new one instead of hoping the problem goes away.
Moreover, the money is following the disruption of tradition. Harvey AI, a legal AI company, reached an $11 billion valuation in February 2026. Just a month later, a rival company named Legora raised $550 million, pushing its value to $5.55 billion– three times what it was worth
five months earlier. These massive valuations reflect real changes on the ground. More companies are using AI to handle legal work themselves rather than hiring outside law firms. In fact, 64% of corporate legal teams expect to rely less on external law firms because of AI.
When put together, that’s the actual shape of AI’s impact on law: automation is rapidly eliminating entry-level grunt work, while unverified AI tools continue to produce fake case law, resulting in severe court sanctions. Thriving practitioners adapt by enforcing strict verification standards and completely rebuilding their training models for new attorneys.
