University AI policies: ten schools, quoted and dated

Each quotation below is the policy's own wording, cut to under 15 words. The link goes to the page the words were read on, not to a homepage or a news report.

Three of the ten are dated on the page itself. Edinburgh's guidance states it is current as of March 2023. Melbourne's statement was last updated on 21 April 2023. Stanford's guidance was adopted on 15 October 2026 and replaces its February 2023 version, and Toronto's FAQ was updated in October 2026. Sydney's page was last updated on 3 November 2026. The older pages have not been withdrawn; they are the policy a student is held to until the university posts a new one.

None of the ten bans AI tools outright across the institution. The difference is where the default sits when a course says nothing.

Stanford's default policy is the most specific of the ten about editing tools. It permits "surface-level editing that does not materially alter the substance, analysis, organization, or argument of the work", with spelling, grammar, punctuation and formatting as its examples, and it says disclosure is not needed for those permitted uses but is required for "any other uses of AI related to coursework" (Stanford Office of Community Standards, generative AI policy guidance, adopted 15 October 2026, read 9 November 2026). A rewrite that changes sentences wholesale is not surface editing under that wording.

Four of the ten address detectors directly, and a fifth institution's teaching centre does. Their positions are not the same, which is why a student should read their own university's page rather than assume.

Toronto's FAQ gives its reason: detection programs "are known to incorrectly flag instances of AI use in human-written content" and are "far less reliable when the output has passed through multiple LLMs, 'humanizer' bots, or is a combination of LLM and human writing" (University of Toronto, generative AI FAQs, updated October 2026, read 9 November 2026). Sydney takes the other path and names Turnitin, whose own FAQ states that its percentage "should not be used as the sole basis for action" (Turnitin, AI writing detection FAQ, read 9 November 2026). No detector score should be the only evidence in a misconduct case, and both the universities that use detectors and the universities that refuse them say so.

What replaces the detector in these policies is a conversation. Stanford states that instructors may ask students to explain and elaborate on their work in class, sections and labs. Toronto recommends meeting the student to discuss the work in person and short in-person assessments that ask students to expand on out-of-class work. UCL names the investigatory viva. The page on how professors check for AI goes through those methods one by one.

The policy tells you whether; the style guide tells you how. MLA asks writers to cite a generative AI tool whenever they paraphrase, quote or incorporate content it created, and to "acknowledge all functional uses of the tool (like editing your prose or translating words)" in a note or in the text (MLA Style Center, citing generative AI, read 9 November 2026). APA asks writers to describe how the tool was used in the Method section or, for a student essay, in the introduction, and to add a reference entry crediting the model's maker (APA Style, how to cite ChatGPT, read 9 November 2026).

UNESCO's guidance on generative AI in education and research, first published in September 2023 and updated in January 2026, asks institutions to set validated rules for GenAI use rather than leave the question to detectors or to students (UNESCO, guidance for generative AI in education and research, read 9 November 2026). The ten policies above are what that work looks like once it is done, and Edinburgh's page adds a practical rule that fits every one of them: acknowledge the tool used, describe how you used it, and give the date you accessed it, even when none of its output appears in your work.

On the generative side of every line these policies draw. Cornell's sample "AI-free" course policy prohibits "tools that help reorganize and edit your written work"; Stanford permits only edits that leave substance, organization and argument untouched; Oxford and Cambridge require any AI contribution to be acknowledged. A humanizer rewrites sentences wholesale, so under all ten policies it is an AI use that must be permitted and disclosed, or not used. The question of whether using an AI humanizer is cheating has its own page, the tools students ask about are listed with their limits on the students page, and the rest of this series sits in the guides hub.