Lawyers Create Legal Puzzles to Train AI Models at Mercor and Micro1

Arbitrator Jessica Crutcher ends her daytime hearings and begins overnight shifts inventing legal disputes for AI models through Mercor. She drafts problems such as an oil and gas trading lawsuit or an asylum case, then evaluates the responses generated by the systems.

Insider reported that Mercor and Micro1 have recruited thousands of lawyers, retired judges, and paralegals to supply professional judgment that public data cannot provide.

More than 1 billion people use chatbots such as ChatGPT and Claude each month. Harrison Margolin began similar work after responding to a LinkedIn post from Micro1. He constructed a cross-border merger exercise with fabricated documents and a scoring rubric, then adjusted the scenario until the models failed to reach correct conclusions.

Mercor lists hourly rates between $100 and $200 for legal experts. Micro1 states that pay depends on seniority and assignment type. Bertie Vidgen, a researcher at Mercor, said legal reasoning remains difficult to teach because much relevant knowledge exists behind paywalls or in confidential agreements.

Entertainment lawyer Charley Kelsey developed an imaginary video game project that included licensing documents and promotional materials to test AI handling of copyright and trademark questions. While analyzing a taxidermied bear copyright issue, he located case law that treats taxidermy as sculpture potentially eligible for protection.

Kelsey left full-time practice at Lionsgate in June to join Micro1.

He said the assignments require slower examination of each reasoning step than typical client work demands. Investors have directed billions of dollars into startups building AI tools for contract review and case law lookup. The lawyers described the compensation as secondary to observing how the technology may alter routine legal tasks.