Loading…
Safety & ethics
Talk handles some of the most sensitive information a person can share. Safety is treated as a foundational design requirement rather than something added once the features work.
Our commitments
Providers submit identity documents and qualification evidence before their profile can go live. An administrator reviews every application. An unverified profile is invisible to patients — not merely flagged.
The AI companion supports; it does not treat. Its scope is explicit, its limits are stated on screen, and escalation to a human is built into the end of every conversation.
Where a conversation indicates possible immediate risk, emergency contacts are shown immediately and synchronously. We do not rely on a queue being watched to keep someone safe.
Journal entries and mood check-ins are visible only to you. Providers see what you choose to share in a session, and nothing else.
Any session, professional or interaction can be reported. Reports are reviewed, and a provider can be suspended from the directory pending investigation.
AI responses are audited for safety, bias and cultural relevance. Safeguarding protocols are developed with clinical input rather than written by engineers alone.
Limits
Being clear about this is part of being safe. A platform that overstates its reach puts people at risk.
Talk is not an emergency service. Nobody is monitoring the platform around the clock waiting to intervene. If you are in danger, the fastest route to help is a phone call to the emergency services, not a message here.
The AI cannot assess risk reliably. Automated detection misses things and over-triggers on others. It is a prompt to surface help, never a clinical judgement.
Verification is not a guarantee of outcome. We check that a provider is who they say they are and holds the qualifications they claim. We cannot guarantee that any particular professional is the right fit for you — and you are free to change.
In immediate danger?
Escalation
The conversation stops. The companion does not continue a wellbeing exercise over a disclosure of immediate danger.
Emergency contacts appear on screen immediately, with tap-to-call numbers.
You are encouraged to contact someone you trust, and offered help finding a professional.
Where appropriate and legally permitted, the case enters a review queue so a human can follow up.
Related
Data minimisation, encryption, retention and your rights — written to be read.