Some educators are trying a different approach to guarding against AI cheating — a “linguistic fingerprinting” technique that borrows a page from the playbook of criminal investigations.