AI and school data

AI in teaching

AI in 321school supports teachers in developing teaching content and feedback. In the school-use model, student AI use is linked to language support and limited help in teacher-led learning activities.

Review draft v0.12. Content revised: November 12, 2026.

School-use principles

  • AI is subject and teaching support. Suggestions may contain errors and do not replace the teacher's judgement.
  • Teachers must check facts, sources, language, tasks and answer keys and approve content before further use, sharing or publication.
  • Suggested teacher feedback must be reviewed, adapted and approved by the teacher before being used as their feedback.
  • Use only relevant information as AI input. Teachers must consider students' age, learning goals and privacy.

Content and quality assurance

Teachers can use AI to develop texts, tasks and other teaching content. Prompts, pasted text and relevant teaching context may be sent to the AI service. Unnecessary student names and private information must be omitted.

  • The text generator includes a review step where teachers approve the text before tasks are created. Content and answer keys must also be checked after task generation.
  • The fact-check tool provides comments and sources for teacher review. It does not replace checks against reliable sources or the teacher's own quality assurance.

Teacher-requested feedback on answers

When a teacher requests AI feedback, student answers, tasks, relevant reading text, results and necessary task context may be sent to OpenAI. A student not chatting directly with AI does not mean their work stays outside AI processing. Free text may contain personal data.

  • This includes teacher-led Space/Lesson work when feedback is requested. Reading results/activity signals may be included; it is not an official diagnosis or automated municipal assessment decision.
  • Teachers assess suggestions against the student's work and learning goals, edit where needed and approve them before using them as their feedback. This is different from teacher-led support given to students during activities.
  • Access to student answers and individual AI feedback requires teaching permission. School administration alone must not grant such access.

Images and audio

Image/writing tools can process prompts, student text and image/task context. Audio/speech tools require feature-specific documentation: storing student recordings and sending data to AI are different questions.

  • Do not assume all recordings go to AI or no media does. The feature annex must describe the text, images or audio actually disclosed. Voices/images may identify students.

Boundary with self-study

Individual self-study is a separate use and is not automatically part of teacher-led school use. This page describes AI for teaching purposes.

  • School-organised work does not become private use merely because a student accepts separate terms. School data must still be processed within the school owner's agreed scope.

Provider, retention and training

321school uses OpenAI Platform/API. The owner reports a DPA signed on 23 October 2026. Account images show all three sharing options disabled, API call logging Enabled per call and Project Residency Global. OpenAI states API data is not used for training without opt-in. Global is neither an EU restriction nor a single processing country. Response storage, Files, prompt caching and security logs have different rules; no general zero-retention guarantee is given. Explicit store:false in the selected text calls was deployed on 28 October 2026. This does not change the other retention rules.

  • OpenAI distinguishes stored responses from security/abuse monitoring logs. Responses normally has a 30-day default retention period or when store:true; in those cases at least 30 days is stated. Security/abuse logs normally remain for up to 30 days, subject to provider exceptions. store:false is not Zero Data Retention. Files, caching and some features have separate rules; there is no single maximum period or zero-retention guarantee. Age-related terms are assessed before use by minors.
  • OpenAI is present in the reviewed AI code. Actual APIs, account/region settings, contract, logs and data must be confirmed for municipal features. Default terms do not establish Fjord Service's account settings.

Approval and publication

Teachers must review and approve content before sharing or publishing. Public release of identifiable student work, images or voices requires separate assessment; the teacher role alone does not justify it.

  • Visibility controls for lesson publishing were deployed and verified in production on 12 November 2026 using synthetic data. Publishing requires an explicit visibility value: public, unlisted or private. The library lists only active public copies. Anyone with the link can open an unlisted copy; this is not private access. Private copies require owner or platform-administrator access and are excluded from public listings and page metadata. Unpublishing closes public access to the copy while retaining its content and the owner's access.
  • Where publishing requires fact-checking, the teacher's review must be confirmed for the current content version. Changes may require renewed confirmation.
  • A confirmed review records the teacher's approval, not an AI-service guarantee that the content is error-free.

Access and production status

Access controls, private file handling and participant-link approval went live on 28 October 2026. Firestore/Storage rules and application flows were checked with synthetic accounts, a dedicated test room and rejection of unauthorised access. Old direct download tokens were revoked for the entire mapped set of 128 audio files, rechecked on 29 October. Files, submissions and memberships were preserved. The school boundary for new invitations was deployed on 6 November and time-limited substitute access on 7 November 2026, both with documented functional checks. The evidence covers the flows and file set examined.

  • Visibility controls for lesson publishing were deployed and verified in production on 12 November 2026 using synthetic data. Publishing requires an explicit visibility value: public, unlisted or private. The library lists only active public copies. Anyone with the link can open an unlisted copy; this is not private access. Private copies require owner or platform-administrator access and are excluded from public listings and page metadata. Unpublishing closes public access to the copy while retaining its content and the owner's access.
  • The approval flow for new account links in Spaces is deployed. A personal participant code alone does not grant a new account/device access to existing participant work. A teacher with the required administrative access approves the link using the participant’s temporary confirmation code. Access can be revoked without deleting work. The flow was checked using synthetic accounts and a test room. Code/QR does not require private email, but does not mean student data is anonymised.
  • The school boundary for new co-teacher and observer invitations to school rooms was deployed and checked on 6 November 2026. Invited teachers must have an active affiliation with the same school. The school administrator role does not itself grant access to individual student answers or the right to invite on behalf of the room owner.
  • Time-limited substitute-teacher access was deployed and checked on 7 November 2026. A school administrator can grant a teacher with an active affiliation with the same school access to a specific school room, including during teacher absence. The substitute can read assignments and student answers and provide written feedback through a dedicated, scoped view. Access can be revoked and expires automatically. Grants, expiry deadlines and revocations are logged. This does not grant ordinary room administration rights or general access to the school’s other rooms.
  • Since 11 November 2026, a platform administrator can disable AI feedback suggestions for Lessons answers and AI help in writing activities across Space rooms linked to one verified school, including new rooms. Rooms may apply stricter settings; a room cannot override the school's restriction. These settings are not automatically enabled for every school. Rooms without a verified school link are outside their scope. Other AI features and school-owner-wide controls are not covered by these two restrictions.
  • Checks using synthetic accounts and test rooms confirmed rejection across the school boundary, continued ordinary student and co-teacher access, and that revocation and expiry block further substitute access. Existing student work was preserved. The evidence covers the flows tested.
  • Planner is paused in the published service. Navigation, direct pages, Planner APIs, AI/calendar entry points and ordinary client access to original plans are blocked. Independent task copies previously shared to Spaces still follow ordinary Spaces access. The pause does not delete or migrate existing content.
  • Production Vercel Functions run in Frankfurt (fra1), verified during deployment on 28 October 2026. The move does not automatically change processing locations for Firebase Authentication, OpenAI, Resend or other services, nor does it confine all logging, CDN, build and support processing to the EEA.
  • Agreed role policy: school administrators manage only their own school's teachers within its licence and receive aggregate school statistics. That role alone must not grant access to individual answers, AI feedback or teaching rooms. Administrators who also teach use separate teaching permissions.

Translation, vocabulary and reading aloud

Translation, vocabulary support and text-to-speech are AI-based support features in 321school. Text for translation or reading aloud and relevant language context may be sent to OpenAI. They can support comprehension, language learning and access to teaching content. Assess them separately from open-ended AI generation and AI feedback. Ordinary student recordings and audio playback do not themselves imply AI processing.

  • Feature and documented status: Translation, vocabulary and reading aloud support teaching. In the model, student access follows the activity and valid room access regardless of personal code, Feide or another login. Text-to-speech sends text and selected voice to OpenAI. Consistent server access and quota enforcement remain follow-up items.

Podcasts and optional AI support

Notes, scripts, participant fields, recordings, segments and finished podcasts may be stored. AI support can send tasks, student text, podcast names, participant fields and other completed work to OpenAI. This does not mean all recordings are sent to AI.

  • Feature and documented status: Teachers control podcast AI support with on/off and usage limits. The server route checks membership and submission binding. Support follows the teacher activity rather than unrestricted student AI. Downloading a finished podcast and public publishing are different actions. Any school-owner restriction on processing student text requires development and testing. Download and export fixes from synthetic tests are included in the published source from 28 October 2026.

Process-oriented writing

Plans, drafts, fields, submissions and feedback may be stored. AI support can send student text, relevant drafts and task/section context to OpenAI. Image support has separate data flows.

  • Feature and documented status: Activity/section AI settings exist. The server route checks whether AI is enabled and which support actions are allowed. These settings are not a unified municipal policy.
  • Possible adaptation for the school owner: For specific requirements, assess one shared restriction on AI support processing student work in writing and podcast workshops. Writing and recordings can remain available without it. Clarify text/media retention together. Shared enforcement requires development and testing.

Responsible use of AI

AI suggestions are working material for teacher review, quality assurance and approval. Data flows, provider processing and any school-owner adaptations are assessed for school use and described in the privacy documentation and agreement.