Skip to content
ChatGPTDetect
English

Can universities detect ChatGPT? What England's regulator said on 11 November

As of 2026-11-12, not reliably, according to England's higher education regulator. On 11 November 2026 the Office for Students (OfS) and Advance HE published Making AI work for students, which states that evidence suggests "AI detection tools do not consistently and reliably identify AI use." The same report says detector output should be, at most, one piece of evidence. That is also how a style-based checker such as ChatGPTDetect should be used.

What the report says about detectors

The OfS and Advance HE ran research between June and September 2026, mainly on English institutions. On detection, three points stand out.

The report is not a blanket dismissal. It notes evidence that some newer detectors may identify AI-generated and modified text more accurately than established tools, but also that gains tend to be temporary and that "humaniser" tools can reduce accuracy. Student participants were strongly critical of detectors, saying they overstate AI use and leave no route to prove innocence after a high score.

What the regulator wants instead

The main recommendation is about assessment design, not detection. Where assessments are unsupervised, the report says, they can often be completed to a high standard with AI. It points to secure formats such as in-person exams, presentations and other controlled assessments. Times Higher Education reports that the OfS tied this to its quality condition B4 and that its director of strategy and delivery, Josh Fleming, called AI a "generational risk to the system as it stands."

What is confirmed

Point Source
Report published 11 November 2026 (ref. OfS 2026.56) OfS publication page
Detection tools "do not consistently and reliably identify AI use" Report, executive summary
Detector output should be at most one piece of evidence Report, paragraph 75
Text watermarking: "the impact of these techniques is not yet clear" Report, paragraph 46
Link to quality condition B4, "generational risk" quote Reported by Times Higher Education

Watermarks do not fill the gap

OpenAI's textGrain watermark applies to ChatGPT and Codex text in the EU only, and OpenAI says the detector is not public. It lists academic-integrity checking of student work as outside its program, and says a missing watermark does not prove human authorship. A university in England cannot rely on it. The details are on the ChatGPT watermark status page, and OpenAI's own educator guidance says detectors have not been reliable enough in its research for judgments about students.

How to use a ChatGPT check responsibly

A style-based detector reads how a text is written and reports a likelihood plus resemblance to model families. It cannot show who typed the words or how much help they had.

More on this workflow is in AI detection for teachers.

Check a text

Paste at least 100 words into the ChatGPT detector for one free check. Veread, the model behind it, wrongly flagged 0.5% of human-written texts in seven languages, as set out on the facts page. The result is writing-style similarity, not proof of authorship, and it works best as one input among several, as the report advises.

Sources: Office for Students and Advance HE, "Making AI work for students: reflections on institutional practice" (11 November 2026); Times Higher Education (11 November 2026); OpenAI Help Center educator FAQ and OpenAI, "Our approach to EU text provenance rules", read on 12 November 2026.

Questions and answers

Can universities reliably detect ChatGPT-written work?

Not by detection tools alone. England's Office for Students said on 11 November 2026 that evidence suggests AI detection tools do not consistently and reliably identify AI use, and that their output should be at most one piece of evidence.

Did the Office for Students ban AI detectors?

No. The report says detector output should not carry disproportionate weight and should not be the basis for a lower mark without strong evidence. It does not prohibit using detectors as one input.

Does the report apply outside England?

The research covered mainly English higher education institutions, and the OfS regulates higher education in England. Other countries and school systems set their own rules.

What should a teacher do with a ChatGPT detector result?

Treat it as a reason to look closer, then compare the text with the student's earlier work and ask them to explain it. A style-based result is a signal, not proof of who wrote the text.