Age-appropriate by default
Interaction, language, permissions and escalation should reflect the learner’s age and context—not rely on one generic chatbot policy.
Safety by design
Trust must be built into every learning interaction.
Because EqualEd is intended for young people, child safety, privacy, accessibility and responsible teaching behaviour are product requirements—not a policy page added after launch.
Designed with leading child-safety and privacy principles in mind. This is not a claim of formal certification or complete legal compliance.
Learnerbest interests
Age context
Curriculum
Teaching policy
Verification
Human review
Six operating principles
The child’s best interests come first.
When commercial, engagement and child-interest objectives conflict, the learner’s safety, rights and development should win.
Minimum necessary data
Collect only what a defined learning, safety or operational purpose requires. A learner’s work is not a free training corpus.
Grounded, controlled teaching
Use approved curriculum material, bounded teaching moves and verification. When the system cannot establish an answer, it should say so.
Human responsibility
Teachers, parents and safeguarding professionals remain accountable. AI can support decisions; it should not quietly become the decision-maker.
No attention business model
No behavioural advertising to children, no sale of student data and no design that pressures learners to stay online for its own sake.
Visible limits and routes to help
Learners should know they are interacting with AI, how to report a problem and when a trusted human is the right next step.
Privacy by restraint
Collect less. Explain why. Keep control visible.
A learner’s free text, work and chat should be treated as their data and potentially their intellectual property—not as a free model-training corpus.
NeverBehavioural advertising to children
NeverSell student data
DefaultMinimum necessary information
RequiredClear access, correction and deletion routes
Defence in depth
Safety is a sequence, not one filter.
- 01Age, role and consent context
- 02Input risk classification
- 03Permitted tools and curriculum retrieval
- 04Bounded teaching policy
- 05Verified model output
- 06Age-appropriate response
- 07Minimal audit and human review
Controlled tutoring
A fluent model is not the learning product.
EqualEd’s intended architecture uses a controlled tutor orchestrator with specialist capabilities. Deterministic systems own permissions, assessment state and calculations; AI is bounded to appropriate teaching tasks.
Ground
Retrieve approved curriculum concepts, examples and rubrics.
Plan
Select one appropriate teaching move for the evidence and objective.
Generate
Use a permitted model with minimum relevant learner context.
Verify
Check maths, citations, schema and safety before responding.
Record
Write only evidence the interaction actually supports.
Transparent limitations
What EqualEd must never pretend to be.
An infallible teacher
A confidential counselling service
A continuously monitored emergency service
A replacement for a school’s safeguarding process
A qualified human making consequential decisions
If serious harm may be imminent
The product should encourage immediate real-world help and follow a documented, jurisdiction-appropriate escalation route. It must not imply that an automated alert guarantees a human response.
Standards-aware, evidence-seeking
Design against the obligations that matter.
The blueprint references UK DfE product-safety guidance, the ICO Children’s Code, COPPA, the EU AI Act and NIST’s AI risk framework. Formal scope, legal and assurance work remains required before launch in each market.
UK DfE guidanceICO Children’s CodeCOPPAEU AI ActNIST AI RMF
Build with us
Safety should be inspectable, not assumed.
EqualEd intends to publish clear product limits, data practices, evaluation evidence and reporting routes before child-facing registration opens.