New research warns that AI writing tutors may deliver less helpful feedback when tools receive students’ protected background attributes. In a Stanford University Institute for Human-Centered AI study, doctoral researchers Mei Tan and Lena Phalen found that providing race, gender, language, disability status, or achievement-related information changed the kind and quality of feedback generated—even when the writing samples were identical. The findings complicate campus adoption of writing-coach tools for accessibility and personalization, raising concerns that “personalization” may reproduce unequal instructional treatment. Researchers presented results at the International Learning Analytics and Knowledge conference. Separately, reporting highlights how AI misuse is reaching beyond cheating to student attempts to defeat AI detectors. As tools become more common in classrooms, the enforcement arms race is intensifying for faculty and administrators responsible for academic integrity. Taken together, these stories signal that assessment governance for AI needs more than tool deployment—it requires validation against bias and clear policy on acceptable use, documentation and review.