NUUBI
Back to Blog

How Nuubi Protects Student Privacy

· Nuubi Team

Privacy pages tend to be full of promises. This page aims to describe what actually happens to a student's data when their course uses Nuubi: what is protected, what is deliberately visible, and which parts depend on the institution. It is written for faculty, IT, and procurement reviewers, and it avoids absolute guarantees, because no vendor can offer them.

Nuubi Works With Course-Scoped Data

Nuubi is a course tool. Students ask course questions in a shared forum (https://www.nuubi.com/blog/graduate-course-ai-forum-case-study), upload drafts for private feedback before submission, work in group chats, and receive coaching on how they use AI. All of that activity produces course-scoped data: questions, posts, drafts, participation, and the course materials faculty provide. Nuubi has no need for a student's wider academic record to do its job.

Course Materials and Student Data Are Not Used for Training

Course materials and student data are not used to train AI models. Nuubi states this in its security documentation, and the backend LLM API is set so that inputs from Nuubi are not used to train the provider's models. A student's draft is used to give that student feedback, and a professor's materials ground answers for that professor's course; neither is pooled across courses or customers. Nuubi's code does not fine-tune or train any model on user content.

What Is Sent to the Backend LLM API

Nuubi's code never adds a username, email address, display name or user id to an AI prompt. In fields that Nuubi fills in, people appear only as role labels such as Admin 1, Member 2 or Student, never as names. Text that people type or upload is sent as written, including posts, comments and messages. Group and course names can appear in prompts, as can internal identifiers such as group UUIDs, post and comment ids, and timestamps on analytics paths. Attachments are sent as up to 6,000 characters of extracted text, not whole files.

Nuubi keeps no log of AI prompts or model responses, and its server logs contain no prompt or model content and no user or group ids. Cuubi answers and support conversations themselves are stored in Nuubi's database.

Visibility in Nuubi Is Designed Around the Course

A course tool deliberately differs from a personal chatbot. Questions in the shared course forum are visible to the whole class, so everyone can learn from each exchange. The instructor can see course exchanges and per-student participation, and instructors also serve as administrators of their course group chats. Draft uploads give each student a working space private from classmates before submission, with feedback tied to the course and its rubric. Reporting stays at the course level, and Nuubi does not produce cross-course reports on students. Moderation reports go to human admins, with no LLM involved in that step.

Student Safety in Nuubi

Several design choices in Nuubi concern student safety in particular.

Identity stays out of AI prompts. Nuubi's code never adds a username, email address, display name or user id to an AI prompt. In fields that Nuubi fills in, people appear only as role labels such as Admin 1, Member 2 or Student, never as names. Text that people type or upload is sent as written.

Course materials stay within their course. Materials a professor provides ground answers for that professor's course only; they are never pooled across courses or customers.

Humans stay in the loop. Moderation reports go to human admins, with no LLM involved in that step. Instructors can see course exchanges and per-student participation, so they can see where students are struggling.

Draft feedback is private. Draft uploads give each student a working space private from classmates, and students see only their own data, enforced on the server.

No public rankings. Nuubi has no public leaderboard and no grade tally.

Protecting Faculty Materials

Course materials get their own set of protections.

Only the instructor can add, see, or remove course materials. Students never receive the materials list or the files. Uploaded course material files are private: they are not reachable by public link, and only the instructor can open them.

Materials stay inside the course. Every read and write is scoped to that course's group, and materials are never pooled across courses.

When a material is deleted, the stored file is deleted too, and it stops being used as AI context at once.

When a material is relevant to a question, it is sent to the backend LLM API as extracted text. The files themselves are never sent, and the text is not used for training.

Students may see excerpts of materials inside AI answers.

AI Training Versus Context

Faculty materials are used as context, not as training data, and the difference matters.

Using context means this: when a student asks a question, the relevant extracted text is included in that single request to the backend LLM API, so the answer can reference the course. Each request is stateless. Nuubi builds the prompt, gets the answer, and discards the prompt; it keeps no log or table of the prompts it sends. The request does not change the model, and what is sent is limited to that request's context, scoped to the course. The backend LLM API is set so that inputs from Nuubi are not used to train the provider's models.

Training means changing a model's weights using data. Nuubi does not do this: no fine-tuning, no training datasets, no embeddings, and no exports of user content for training. Because a request does not change the model, course content does not become part of the model or appear in answers for other courses or customers.

Faculty stay in control of the context itself. Instructors add and remove materials, and deleting a material stops its use as AI context immediately.

The same limit stated earlier still applies: text that people type or upload is sent as written, so personal details do not belong in the text of posts or files.

What keeps course content contained:

1. No training or fine-tuning happens on Nuubi content, and the backend LLM API is set so that inputs from Nuubi are not used for training, so the model does not learn from course content.

2. Each request is separate. Nothing is carried from one request to the next.

3. Materials are pulled only from that course, never from another course.

4. Only extracted text is sent, not the files themselves.

Human Expert in the Loop

AI answers in a course benefit from a subject matter expert who can see them. In Nuubi, the instructor can see course exchanges and per-student participation, which lets the faculty member step in when an AI answer is incorrect or incomplete, and students can see that correction.

Major education bodies and universities call for human oversight of AI in education. The OECD's Digital Education Outlook 2023 treats AI tools as more trustworthy when framed with a human-in-the-loop approach (https://www.oecd.org/en/publications/oecd-digital-education-outlook-2023_c74f03de-en/full-report/opportunities-guidelines-and-guardrails-for-effective-and-equitable-use-of-ai-in-education_2f0862dc.html), and UNESCO's guidance on generative AI in education recommends a human-centred approach (https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research). Universities give the same direction: the University of Minnesota tells instructors that human review of AI output is essential and that AI supports instructor judgment rather than replacing it (https://teachingsupport.umn.edu/teaching-generative-ai/assessing-student-work-genai-tools), and the University of Rochester calls human oversight essential, with instructors monitoring a course AI tool's outputs (https://www.rochester.edu/provost/gen-ai-education/). This kind of visibility, paired with faculty verification, is often not built into general-purpose AI tools, where an answer reaches the student with no expert in the loop.

US Hosting and an Eight-Year Default Retention

Nuubi is hosted on US cloud infrastructure. Data is retained for eight years by default, and an institution can ask for a shorter retention window or earlier deletion for its courses.

Institutions Can Review Nuubi's Security and Accessibility Documents

Nuubi states FERPA compliance publicly (https://www.nuubi.com/blog/student-data-ai-tools-ferpa-questions). A completed HECVAT 4 and a VPAT 2.5Rev reflecting WCAG 2.2 AA conformance are available to institutions on request, which gives a review team detailed answers to work from. To request either document, email support@nuubi.com.

Limits: What Nuubi Does Not Filter

One limit is worth stating plainly: no tool can prevent a student from pasting something sensitive into a text box, so people should avoid putting personal details in the text of posts or files. Text typed in posts, uploaded files, screenshots and course materials are sent to the AI as written. Nuubi's protections cover the fields it fills in; they do not rewrite what a person types. Where Nuubi controls the field, minimization limits what any mistake could reach.

Further Reading

For the full checklist version of the whole-system argument, see Beyond the Model: AI Privacy, Safety, and Security for Students (https://www.nuubi.com/blog/ai-privacy-safety-security-students). For a comparison of Nuubi's posture with Claude, ChatGPT, and Gemini on a single checklist, see AI Tools With Student Privacy Protections: Nuubi, Claude, ChatGPT, and Gemini Compared (https://www.nuubi.com/blog/ai-tools-student-privacy-protections-compared). Reviewers are welcome to send questions to support@nuubi.com.

Sources

Nuubi: https://www.nuubi.com

Nuubi Technical Security Summary (available to institutions on request; includes HECVAT 4 and VPAT 2.5Rev dated 2026-09-10)

Nuubi, Beyond the Model: AI Privacy, Safety, and Security for Students: https://www.nuubi.com/blog/ai-privacy-safety-security-students