How this helper keeps you safe
This is a signposting service powered by AI. We've put six concrete protections in place so the AI can't invent advice, can't replace a real person, and can't keep you in a loop when you need a human. Here they are, in plain English.
The AI can only point you to organisations we've already checked.
Behind the scenes we keep a verified directory of UK Muslim-led mental-health services — every helpline, charity and counsellor on it has been reviewed. The AI is only allowed to recommend things from that list. If it ever tries to invent a name, our system strips it out before you see the reply.
No invented helplines. No made-up phone numbers.
Crisis messages never reach the AI.
When someone types words that suggest immediate danger — for example "I want to die" or "I'm going to hurt myself" — the message is intercepted before the AI sees it. You're shown Samaritans (116 123), SHOUT (text 85258), NHS 111 and 999, straight away. We don't risk an AI answer at the moment you most need a human.
This helper does not diagnose, treat, or prescribe.
We point you toward people who can. The AI gives you context, suggests verified services, and — when you ask about Islamic framing — quotes from material already reviewed by scholars. That's it. For diagnosis, treatment, or therapy, you'll be signposted to a real clinician or counsellor.
When the AI isn't confident, it says so.
If your question doesn't match anything in our verified directory closely enough, the AI doesn't guess. It says "I don't have a confident match" and points you to the full directory to browse yourself. We'd rather show you nothing than show you something we're unsure about.
Your conversation is kept for safety auditing — and you can ask us to delete it.
We keep an anonymous log of each conversation for up to 12 months. That log lets us check the AI is behaving — it's how we catch problems before users do. We never sell this data, never share it with advertisers, and we never link it to your name.
You have a legal right (UK GDPR) to ask us to delete your conversation at any time. Email us with your approximate date and time and we'll erase the record.
A real person reviews how this helper is performing every week.
Every Friday we run 50 challenging test questions through the helper and check the answers. Bad answers go on a fix-list before the next deploy. We're a small team — if something feels wrong, write to us; we read every message.
Email [email protected] — please include the rough date and time so we can find the conversation. We aim to reply within five working days.
For an urgent mental-health concern, please use the crisis lines at the top of this page.
For the technically curious
The six promises above are the user-facing summary of a portfolio-wide safety posture we apply to every AI service the Muslim Council of Britain ships. The technical detail lives in two places:
- Our
AIMSdocument — what governs this AI helper, who's accountable, what we measure, when we review. - Our shared
hallucination-mitigation playbook— the four-layer technical pattern this helper implements.
We don't link these from the live site yet — they're in the codebase repository (planned to move to a public GitHub mirror once we have the cycles). Email [email protected] if you'd like a copy before then.
References shaping our approach: EU AI Act Article 50 (transparency, 2 Aug 2026), MHRA Software-as-Medical-Device guidance, ISO 42001 (AI Management Systems), and the 2026 academic consensus on retrieval-augmented generation safety.