AI Ethics & Safety

Why this page exists

AnvyoAI uses AI systems to generate recommendations, summaries and mentor responses for school students. That deserves a plain explanation of what the technology does, what it cannot do, and where a human belongs in the process.

Where AI is used

• Summarising discovery activity responses into readable profile signals
• Suggesting career fields, academic pathways and study options to explore
• Drafting and updating personalised roadmaps
• Answering student questions through the AI mentor
• Explaining what a career, exam or programme involves in plain language

What AI does not do here

• It does not make a final decision about a student's education or career
• It does not diagnose any condition, or assess mental or physical health
• It does not replace teachers, counsellors, parents, psychologists, doctors or any qualified professional
• It does not decide admission, selection or eligibility for anything
• It is not an emergency service and must not be used for medical, psychological, legal, financial or crisis situations

Known limitations

These limitations apply to current AI systems generally, and to AnvyoAI.

• Errors stated confidently. AI output can be wrong while sounding certain.
• Stale information. Exam dates, eligibility rules, fees and programme details change and may not be current.
• Gaps in coverage. Less common careers, regional programmes and newer institutions may be described thinly or not at all.
• Bias. Training data reflects patterns in the world, including patterns we would not want reproduced in career guidance.
• Context limits. The system knows only what it has been given, and no student's circumstances fit entirely into a profile.

Human oversight

The platform is built so that a person remains in the decision.

• AI-generated guidance is presented as possibilities to explore, using language such as "worth exploring" rather than "you should become"
• Students are prompted to verify consequential information against official sources
• Students are encouraged to discuss directions with a teacher, counsellor, parent or guardian
• Users can request human review of a profile summary or recommendation they believe is wrong

AI providers

Third party AI providers used: [TO BE COMPLETED BY SHL]

Regions where processing occurs: [TO BE COMPLETED BY SHL]

This section must list the actual providers in use before publication, and must be updated whenever a provider changes. Do not publish a provider name that is not contractually in place.

Your information and model training

Current position on model training: [TO BE COMPLETED BY SHL]

This statement must reflect what is technically and contractually true, including the terms agreed with each AI provider. A claim that user data is never used for model training may only be published if it is accurate for every provider in the processing chain.

Bias and fairness

Career guidance can carry assumptions about who belongs in which field. We work against that in the following ways.

• Recommendations are framed around demonstrated interests and aptitude signals
• Field descriptions avoid assumptions about who typically enters them
• Exploration deliberately surfaces fields beyond the obvious match, so options widen rather than narrow
• Reported cases of biased or inappropriate output are reviewed and used to adjust the system

Safety boundaries

The AI mentor is scoped to academic and career guidance. Where a conversation moves towards medical, psychological, legal, financial or crisis territory, the intended behaviour is to decline to advise and to direct the student towards a qualified person or an appropriate service.

If you are in immediate danger, contact local emergency services. If you are struggling with your mental health, speak to a trusted adult or a qualified professional.

Reporting a problem

Report inaccurate, biased, unsafe or inappropriate AI output to [TO BE COMPLETED BY SHL]. Include the feature, roughly when it happened and what the output said. Reports are reviewed and used to improve prompts, data and safeguards.

Review of this page

This page is reviewed whenever AI providers, model configurations or safeguards change. The effective date at the top reflects the current version.