University of Cambridge AI Policy: How Digital Education Guidance Shapes Coursework in 2026

What Does the University of Cambridge AI Policy Actually Say About Digital Education?

Cambridge's approach to generative AI in teaching and assessment is deliberately decentralized. Rather than issuing one binding rule that applies identically to a physics problem set, a history essay, and a computer science dissertation, the university's digital education guidance sets out shared principles — academic integrity, transparency, and appropriate acknowledgment of sources — and then delegates the specific permitted-use rules to individual faculties, departments, and course organizers. This means a student in one Tripos subject may be explicitly permitted to use a large language model for brainstorming or code drafting, while a student in another department may face a complete prohibition on any AI-generated text appearing in submitted work. The guidance frames this flexibility as necessary because the risk profile of AI assistance differs enormously by discipline: a generated paragraph poses a different integrity question in a reflective essay than a generated function poses in a programming assignment. What stays constant across every faculty is the expectation that students check the specific rules for each course they are taking rather than assuming a blanket policy applies, and that any use of AI tools beyond what is explicitly permitted must be disclosed rather than hidden. Supervisors and directors of studies are typically the first point of reference when a student is unsure whether a particular use case — using AI to summarize a reading, to check grammar, to suggest an outline — falls inside or outside what their course allows.

"We are not trying to write one rule that fits every subject. What has to be consistent is that students know where to check the rule for their own course, and that undisclosed use is treated as a breach of trust regardless of the subject." — Cambridge digital education guidance summary, paraphrased from published faculty briefings

Why Does Cambridge Delegate AI Rules to Individual Departments Instead of Setting One Policy?

The decentralized model reflects how Cambridge's collegiate and Tripos-based teaching structure already operates for most academic matters — assessment formats, marking criteria, and permitted resources have long varied by faculty rather than being centrally standardized. Extending that same logic to generative AI kept the new guidance consistent with existing academic governance rather than creating a parallel, university-wide bureaucracy. There is also a practical reason: the pace of change in AI tools has been fast enough that a single detailed rulebook risked becoming outdated within a single academic year. By setting principles centrally and delegating specifics, individual departments can update their own guidance as tools change — a computer science department can revise what counts as acceptable AI-assisted coding help as new tools emerge, without waiting for a university-wide review process. The tradeoff is that this flexibility puts more responsibility on students to actively find and read the relevant guidance for each course rather than relying on a single memorized rule. Directors of studies and course handbooks are the primary channels through which this information reaches students, and course syllabi increasingly include an explicit AI use statement alongside existing academic integrity language.

1. Central digital education guidance sets shared principles: integrity, transparency, disclosure
2. Individual faculties and departments define specific permitted and prohibited uses
3. Course handbooks and syllabi are the primary place students find discipline-specific rules
4. Directors of studies and supervisors clarify edge cases not explicitly covered
5. Departmental guidance can be updated independently as new AI tools become available

How Should Students Disclose AI Use Under Cambridge's Digital Education Guidance?

Where a course permits some level of AI assistance, Cambridge's guidance generally expects disclosure that mirrors how students already acknowledge other forms of help, such as citing a source or crediting a collaborator. In practice this means noting which tool was used, for what specific task — drafting an outline, checking code syntax, generating a first pass at a literature summary — and what was subsequently changed or written independently by the student. A brief statement at the end of an assignment or in a footnote is the most common format departments ask for, similar to an acknowledgments section in a dissertation. The level of detail expected scales with how central the AI-assisted portion is to the final submission: a student who used a chatbot to check grammar on a finished essay faces a lighter disclosure expectation than one who used AI to generate a substantial portion of analytical content that was then edited. Silence is the core problem the guidance targets — not the technology itself. A student who uses a permitted tool but does not disclose it, or who discloses a lighter form of assistance than what was actually used, is treated as having misrepresented their submission regardless of whether the underlying assistance would have been allowed if disclosed honestly. This is why Cambridge's framing consistently separates two distinct questions: was this use of AI appropriate for this course, and was it disclosed accurately — a submission can fail on either question independently.

1. Identify whether your specific course permits any AI assistance before starting an assignment
2. Note the exact tool and task if assistance is used — drafting, summarizing, code-checking, editing
3. State clearly what was AI-assisted and what was written or verified independently
4. Match the level of disclosure detail to how central the AI-assisted portion was to the final work
5. When uncertain whether a use case needs disclosure, ask your supervisor or director of studies before submitting

How Are Cambridge Courses Verifying That Submitted Work Follows the AI Policy?

Verification approaches vary by department in the same way permitted-use rules do, but several common methods have emerged across Cambridge faculties since generative AI guidance was introduced. Many supervisors rely on direct knowledge of a student's writing and reasoning style built up over regular supervisions, making an unexplained shift in vocabulary, argument structure, or technical fluency a more immediate signal than any automated tool. Written examinations and in-person vivas — already central to Cambridge's assessment culture — have taken on additional importance as a way to confirm a student can explain and defend the reasoning behind submitted coursework, since a student who cannot account for their own argument under questioning raises concerns independent of any detection software. Some departments have also begun incorporating AI detection tools as a supporting signal rather than a sole determinant, consistent with the university's broader emphasis on human academic judgment over automated verdicts. A detection score, where used, functions as a prompt for closer manual review rather than as evidence on its own — a pattern consistent with how most UK higher education institutions have positioned these tools given documented false-positive risks in automated AI detection. Students preparing a submission for a department that has signaled it may check AI-assisted content — whether through direct conversation, a viva, or a supporting detection tool — benefit from being able to explain their own drafting process and, where relevant, confirm that their writing does not carry stylistic patterns that would create ambiguity about undisclosed AI involvement.

1. Supervisor familiarity with a student's usual writing and reasoning style remains a primary check
2. Vivas and oral examinations allow direct confirmation that a student can explain their own submitted reasoning
3. Some departments use AI detection tools as a supporting signal, not a standalone verdict
4. A flagged score typically triggers closer manual reading rather than automatic penalty
5. Students should be prepared to walk through their drafting process if asked

"A viva question about your own argument tells us more in two minutes than any score does. The tools are useful for flagging where to look more carefully — they are not the finding itself." — paraphrased from Cambridge faculty guidance on assessment integrity

What Happens If a Student's Work Is Flagged Under Cambridge's AI Guidance?

Because permitted-use rules and verification methods sit at the departmental level, so does the process that follows a flagged submission. In most cases, a supervisor or examiner who has concerns about undisclosed AI use will first raise the matter directly with the student — through a conversation, a request for drafts or notes, or a follow-up question during a supervision — before any formal academic misconduct process begins. This mirrors Cambridge's long-standing approach to suspected plagiarism, where informal resolution at the course level is the norm and formal referral to a college or university body is reserved for cases involving stronger evidence or a pattern across multiple pieces of work. Students who can produce drafts, notes, or a clear account of their research and writing process are generally better positioned to resolve an informal query quickly. Where a case does proceed to a formal academic misconduct process, the university's existing procedures — the same ones used for plagiarism and other integrity concerns — apply, and a student has the right to respond and present evidence. Given the decentralized nature of the policy, students facing a query about AI use are encouraged to review their specific course or department's published guidance as a first step, since the exact permitted-use boundary they may have crossed — or not crossed — is defined there rather than in a single central Cambridge-wide document.

How Can Students and Researchers Check Their Own Work Before Submitting at Cambridge?

Given that disclosure accuracy matters as much as the underlying use of AI itself under Cambridge's digital education guidance, a practical habit for students is reviewing a draft before submission to confirm it reflects what they intend to disclose — and, separately, confirming that any independently written sections do not carry stylistic patterns likely to invite unnecessary scrutiny. Tools like NotGPT let students and researchers paste a draft and see a sentence-level breakdown of AI-likeness, which is useful less as a pass-fail gate and more as a way to catch passages where heavy editing or a narrow, formal writing style might read ambiguously to a supervisor or examiner. This is particularly relevant for research students writing in a technical register, or for students writing in English as an additional language, both of which can produce prose that reads as more uniform than typical native informal writing without any AI involvement at all. Running a self-check several days before a deadline — rather than the night before — allows time to revisit specific flagged passages, confirm the disclosure statement matches the final draft, and, where a course permits limited AI use, make sure the acknowledgment describes what was actually done. None of this replaces reading the specific guidance published by a student's own faculty or department, which remains the authoritative source for what is and is not permitted in a given course.

1. Read your specific course or department's published AI guidance before starting an assignment
2. Draft your disclosure statement alongside your work, not as an afterthought before submission
3. Run independently written sections through an AI detector to catch ambiguous stylistic patterns
4. Revise flagged passages for natural sentence variation rather than uniformly formal phrasing
5. Confirm your final disclosure statement matches exactly what assistance was used
6. Keep drafts and notes in case a supervisor asks about your writing process