AI Transparency Statement

What Mrs J Sees

When you chat with Mrs J, she has access to:

• Your current message: What you're asking for right now
• Conversation history: Recent messages in your current chat (if context memory is enabled)
• Your preferences: Tone, response length, curriculum, and document format settings
• Teaching context: Key stages and subjects you've told us you teach
• Your materials: Resources you've uploaded (templates, branding, curriculum frameworks) - only when relevant

How Mrs J Creates Content

When you ask Mrs J to create something, here's what happens:

1. Understanding your request: Mrs J analyses what you're asking for, including the document type, subject, year group, and any specific requirements.
2. Applying your preferences: She uses your settings (tone, length, curriculum) to tailor the output to your style.
3. Generating content: Mrs J uses a suite of UK-hosted AI models, all served by relax.ai (a Civo Ltd trademark) on Civo LON1 — one legal entity, one DPA, one UK jurisdiction. Microsoft Azure OpenAI Service (UK South) is the automatic fallback if any relax.ai path is unavailable.

   Models in use and their roles (last verified May 2026):

   • Llama 4 Maverick (17B×128E) — primary generation: chat, lesson plans, slides, worksheets
   • DeepSeek-V4-Pro — agentic tool-calling: calendar, email drafts, external integrations
   • Kimi 2.6 — agentic tool-calling: alternate reasoning mode
   • Mistral-7B-Embedding — semantic search and curriculum retrieval (embeddings)
   • Voxtral — voice transcription and audio responses
   • Azure OpenAI gpt-4.1-mini / text-embedding-3-large — automatic fallback for generation and embeddings

   Honest note on content safety architecture:

   Content safety on the primary relax.ai path is application-layer: a constrained educational system prompt, intent classification, hard routing rules the model cannot override, and a two-pass self-critique pipeline that rewrites outputs failing internal quality checks. All external actions (sending an email, writing to a calendar) require explicit teacher approval before execution. There is no dedicated third-party moderation API on the relax.ai path. Azure’s own content filtering is active on the Azure fallback path.
4. Quality checks: Content is structured according to UK curriculum standards and educational best practices.
5. Delivery: You receive the content to review, edit, and use in your teaching.

How Mrs J Learns

With AI Learning enabled:

• Mrs J remembers your teaching style and preferences across sessions
• She learns which formats and structures work best for you
• Content quality improves over time based on patterns in your requests
• Anonymised insights help us improve the service for all teachers

With AI Learning disabled:

• Mrs J still uses your preferences for the current session
• Each session starts fresh without personalisation from past interactions
• Your conversations are not used for service improvement

You can change this setting at any time in Settings → Privacy & Data.

Data Protection

Your data is protected by:

• UK data sovereignty: all customer and personal data is processed within the UK on Civo LON1 (primary) and Microsoft Azure UK South (fallback), pinned by configuration
• Encryption: data is encrypted in transit (TLS 1.3) and at rest (AES-256)
• Access controls: only you (and, where the school is the controller, your school administrator on the terms set out in our Privacy Policy) can access your conversations and generated content
• Binding no-training commitment: we do not use your prompts, conversations, or generated content to train AI models — not our own, not Civo's, not relax.ai's, not Microsoft's. This commitment is binding and is reflected in our Data Processing Agreement with schools. If we ever offer a UK-trained-model contribution feature in the future, it will be a named, per-school opt-in with a separate consent flow and the default off; pre-existing prompts and outputs will never be retroactively pulled into training.
• UK GDPR, DPA 2018, DUAA 2025, and PECR compliance: see our Privacy Policy

What Mrs J Won't Do

Disclosure to Pupils and Parents

When a teacher uses a worksheet, slide deck, or piece of feedback that Mrs J helped draft, we recommend a one-line credit to pupils and parents. UNESCO recommends it. The EU AI Act's limited-risk transparency principle (Article 50) recommends it. The Australian Framework for Generative AI in Schools recommends it. We are on the right side of the consensus.

We recommend, we do not mandate. Disclosure decisions sit with the teacher and the school, in language and at a level suited to their community. The school AI policy template we publish (alongside our Trust Hub) sets out the recommended wording, and we collaborate with adopting schools on the announcement to parents so the relationship between the school and its community is mediated by the school, in the school's brand, not by us.

Schools that need a starting-point policy in the next ten minutes can use our free generator: Mrs J drafts a UK-specific staff-use policy live, mapped to UK GDPR, DUAA 2025, the DfE GenAI Product Safety Standards, KCSIE 2025 and JCQ, and emails the PDF and Word version straight to your DPO for review and governing-body sign-off.

Regulatory Framework

This statement is written against the following regulatory and policy framework, all of which we monitor and re-check on the day we publish anything that relies on them:

• UK GDPR, the Data Protection Act 2018, and the Data (Use and Access) Act 2025 — our primary lawful basis framework
• EU AI Act (Regulation (EU) 2024/1689) — Article 50 transparency principles inform our recommendation on disclosure to pupils and parents; our deliberate decision not to mark pupils' work keeps Mrs J out of Annex III "education" high-risk territory
• DfE Generative AI Product Safety Standards (January 2026 revision) — the UK ceiling we design Mrs J to clear; we publish a point-by-point LESSO mapping to each Standard for procurement teams
• UNESCO Guidance for Generative AI in Education and Research — informs our human-in-the-loop and disclosure-recommendation positions
• ICO Children's Code (AADC) — the standard we point to when we explain why pupils never log in to LESSO

Context Memory

When "Remember conversation context" is enabled (the default), Mrs J can:

• Reference earlier messages in your conversation
• Understand follow-up requests like "make it shorter" or "add more examples"
• Build on previous work without you repeating yourself

If you disable context memory, each message is treated independently. You can change this in Settings → Mrs J Preferences.

Limitations

Please be aware that:

• AI-generated content should be reviewed before use with students
• Mrs J may occasionally make mistakes or include outdated information
• Complex or unusual requests may not produce ideal results on the first try
• Mrs J cannot access the internet or real-time information unless you specifically ask her to search

We're constantly improving Mrs J's capabilities, and your feedback helps us get better.

Your Controls

You have control over how Mrs J uses your data:

Questions?

General questions: hello@lesso.co.uk.

Privacy and data-protection matters: Luke, our co-founder, handles all compliance and can be reached directly at support@lesso.co.uk.

For more details about data protection, see our Privacy Policy.