New research and policy scrutiny show that generative AI use in education is raising both academic integrity and fairness concerns. Stanford researchers found that AI writing coaches changed the quality and type of feedback when the tool was given student race, gender, language, disability status, or motivation—despite using identical writing samples. The study, presented at the International Learning Analytics and Knowledge conference, examined feedback across multiple model families including GPT-4o, GPT-3.5-Turbo, and Meta’s Llama models. Researchers said teachers may not realize how model-based “personalization” can produce uneven support, raising questions for assessment design and responsible deployment. Meanwhile, student anxiety is escalating as institutions tighten AI detection and misconduct reviews. An Australian survey of more than 1,200 students reported high levels of stress driven by fears of wrongful cheating flags, breaches of privacy, and unfair grade competition, highlighting the institutional response as a key factor in student experience.