Spain has no nationwide law governing how university students may use generative AI tools such as ChatGPT, Gemini or Claude in coursework and exams. The rules are set by each institution and, in practice, by each lecturer through the course syllabus. Yet the official guidance that does exist converges on three principles: AI is not banned by default, using it without declaring it is the fastest route to academic misconduct, and an AI detector does not constitute proof against anyone.
- No blanket ban: Spanish universities treat generative AI as a permitted tool when its use is authorised and disclosed.
- Disclosure is mandatory: Students are expected to document which tool, version and prompts were used, and what output was incorporated.
- Detectors are not evidence: Institutional guidance states that AI-detection reports are indicative at best and cannot alone support an accusation of fraud.
A regulatory vacuum filled by institutional guidance
As of November 2026, no state regulation defines what a student may do with generative AI in an assignment or an exam. The Ministry of Science, Innovation and Universities has opened a public consultation, running from 12 November to 2 December 2026, on a draft royal decree that amends four university regulations and references artificial intelligence. However, the text addresses something else: it obliges each university, unless its autonomous community has its own body, to create an ethics committee that sets criteria on the use and limits of AI in faculty research activity, under a new article 19 of Royal Decree 640/2021. Beyond a generic mention in the preamble, it says nothing about exams or coursework, and it may still change.
The association of Spanish university rectors, CRUE, has not imposed common rules either. Its March 2024 document on generative AI in university teaching explicitly asks each institution to draw up its own guidelines. At the fifth CRUE-Docencia conference, held at Universitat Jaume I on 5 and 6 March 2026, the conclusions of the AI roundtable repeated the same message: assessment methods must be redesigned and responsible use must be taught. The answer to whether AI is allowed in a given assignment therefore lies in the course syllabus, not in legislation.
Permitted support versus outright fraud
The guidance reviewed separates assistance from substitution. The UNED, in its academic integrity guide for final degree projects published on 10 December 2025, makes clear that its position is not prohibition. With the supervisor’s authorisation and a mandatory declaration, it accepts uses such as generating initial ideas, translating to understand a text, checking grammar and style, suggesting database search strategies or adapting references to a citation format.
On the fraud side, it lists presenting AI-generated text as one’s own, including generated graphics or code without disclosure, citing works never consulted that an AI supplied, simulating data, and delegating precisely the task being assessed, such as interpreting results or writing conclusions.
The University of the Basque Country (EHU), in its principles and recommendations on generative AI published on 17 September 2025, advises students to use it for summarising, building concept maps, reviewing texts or generating self-assessment questions. It prohibits presenting its content as one’s own, using it without declaring it, and deploying it in exams or continuous assessment where it has been explicitly banned. Its FAQ adds a frequently overlooked misuse: entering personal or confidential data into AI tools without adequate protection.
The University of Granada (UGR), in a non-binding summary of recommendations drawn up with 108 participants, compares AI to a calculator: using it is not fraud in itself unless the subject prohibits it.
The practical test for students is not whether they will be caught, but whether the task assesses exactly what they are asking the machine to do. If the answer is yes, it is not help; it is submitting an exam completed by someone else.
How to declare and cite AI use
With a dedicated section and citations where needed. The UNED requires final degree projects to include an annex titled Declaration of use of generative artificial intelligence, alongside a sworn declaration of authorship. The annex should record four things: the tool and its version, the most relevant prompts, which outputs were used, and a reflection on what was modified and what the student contributed.
If a generated fragment appears in the body of the text, the UNED refers students to style manuals such as APA, MLA and Chicago to cite it like any other source. The EHU specifies the format in its annex I, based on APA citation recommendations: generated text goes in quotation marks with a citation such as (OpenAI, year), and the bibliography records the tool, version, model type and either a link to the shared conversation or an annex containing the transcript.
The EHU distinguishes three uses: generative (drafting or rewriting), auxiliary (translating) and consultative (correcting style, finding information or analysing data). For the last, a prompt alone is insufficient; the process must be described in the methodology.
If an AI supplies a bibliographic reference, verify it before citing it. The UNED classifies fabricated references as fraud, and there are straightforward steps to check whether a study, citation or law produced by an AI actually exists.
Why a detector cannot prove misconduct
The universities say so in writing. The UNED guide states that no tool today can determine with certainty whether a text was written by AI and that its reports are not conclusive evidence, so they cannot ground an accusation of academic misconduct. Even for Compilatio, the software integrated into its platforms, it notes that results are indicators requiring lecturer judgement.
The EHU goes further, citing research showing that detectors have a high false-positive rate and that lecturers’ intuition performs even worse. The UGR explains that these tools only produce a probability and that each model improvement makes them less reliable. CRUE itself argued in 2024 for focusing effort on prevention rather than detection.
So how is authorship verified? Through staged submissions and, above all, an oral defence with unscripted questions, according to the UNED. The EHU proposes giving greater weight to in-person assessments than to the submitted document.
For students, the direct consequence is to keep drafts, notes and the document’s version history. If an accusation ever rests on a detector score, that record is what demonstrates the work is theirs.
Spanish higher education has arrived late but at a sensible conclusion: banning what cannot be detected does not work, while demanding transparency and assessing the process does. The real risk is fragmentation. With every faculty and subject setting its own limit, a student can end up facing disciplinary action for something permitted in the course next door. When in doubt, ask in writing and always declare. An extra declaration has never failed anyone.