Lawyers turn to AI training as a new revenue stream

Published: 2026-07-03T13:39:42.000Z

For a growing number of lawyers, the most interesting billable hours of the week now happen after dinner, and the client is a machine. Legal professionals are taking paid evening work training AI models by inventing hard problems, prompting the systems and grading their answers.

For a growing number of lawyers, the most interesting billable hours of the week now happen after dinner, and the client is a machine. Business Insider has reported that legal professionals, among them the arbitrator Jessica Crutcher, are taking paid evening work training AI models by inventing hard legal problems, prompting the systems and grading their answers. It is a quietly booming side hustle, and for many it has become a serious second income stream at a moment when the profession is nervous about what artificial intelligence will do to the day job. The work is funneled through specialist talent platforms that sit between the AI labs and the experts. Vendors such as Mercor and Micro1 have enlisted thousands of lawyers, retired judges and paralegals for this kind of task, driven by the fact that more than a billion people now use chatbots like ChatGPT and Claude every month and expect legally fluent answers. The logic is straightforward enough. Scraping the open web only gets a model so far, and closing the reliability gap on nuanced statutory or adversarial questions takes genuine domain judgement that only a trained lawyer can supply. Mercor has become the poster child for the trend, and the numbers are eye-catching. The San Francisco platform supplies white-collar talent to labs including OpenAI, Anthropic and Meta, and according to a Bloomberg report it pays out more than two million dollars a day to its contractors. Its network runs to roughly 30,000 doctors, lawyers, bankers and writers, with contributors earning an average of around 85 dollars an hour, though verified legal specialists can command far more. There is an obvious irony that few participants miss, since many are being paid to teach a system that may one day absorb parts of their own work, and some have complained of heavy surveillance and being treated like gig labor rather than professionals. The regional dimension is only going to grow. The Gulf has thrown enormous resources at building its own AI champions, from Saudi Arabia's HUMAIN and its Arabic-first models to the UAE's G42, and those systems need to reason accurately about Sharia principles, GCC commercial codes and the common-law frameworks that govern the DIFC and ADGM courts. That is expertise no Western dataset holds, which makes lawyers versed in Saudi, Emirati and wider regional law valuable trainers as the platforms recruit internationally. With Vision 2030 pushing digitization across government and business, and legal AI adoption climbing fast in the region, the same after-hours economy that is paying American attorneys looks set to reach the Middle East's legal community before long.