Maya AI · Responsible AI
Trust isn't a policy page. It's the product.
People talk to Maya at the moments they have most to lose by talking: a resignation, a bad first month, a restructure they were told was final. They only do it because they know exactly where their words go. So confidentiality is not a promise made afterwards. It is how Maya is built, from the first hello to the final report.
Trusted Secured Without Bias Never linked to a nameYour people's trusted confidante
Commissioned by the organisation. Never reporting on the person.
What Maya does, and never does
Six rules that do not bend for a client.
A response is never linked to a name
Maya's records hold no employee ID, no phone number and no name against what was said. There is no lookup table, because there is nothing to look up.
Nothing is reported from a small group
Groups under five people cannot be invited, and a program that does not reach five completed conversations closes without a report. We would rather tell you nothing.
No manager reads a raw conversation
Recordings and transcripts are used to identify themes and are never handed to anyone inside your organisation. Quotes in a report are paraphrased and stripped of identifying detail.
Maya says what she is doing, up front
Every conversation opens with what Maya is, who commissioned her, and what happens to the words. People can stop at any point, and stopping has no consequence.
Taking part is always voluntary
Nobody is required to talk to Maya, and non-participation is never reported at an individual level.
Maya is not a monitoring tool
She will not be used to assess, rank or investigate an individual. If a client asks for that, the answer is no, and it is in the contract.
Humans stay in the loop
Maya runs the conversations. Our researchers review how she asks, check the themes she draws, and sign off every report before it reaches a client. No finding goes out on the model's word alone.
We test her questioning for leading language and her analysis for bias across demographics and languages, and we retire phrasing that skews an answer. When Maya is not confident a theme is real, the report says so instead of dressing it up.
A recommended action always carries the weight of evidence behind it: nine of twenty-seven people, not "a concern was raised".
Data, handled carefully
Contact details used to invite people are held separately from anything anyone says, and are deleted when a program ends. Conversation data is encrypted in transit and at rest, and retained only as long as a program needs it.
Your data is never used to train a public model. Access inside our team is limited to the researchers working on your program.
If there's real risk
One careful exception, stated openly.
If someone tells Maya they are in danger, Maya says clearly that this is one thing she cannot keep to herself, points them to real help, and asks their consent before anything is escalated through the route agreed with your organisation. People are told this before they start talking, not after.
Enterprise security
Built for organisations that cannot afford to get this wrong.
ISO 27001 Certified · EU GDPR Compliant · AWS Enterprise Cloud.
ISO 27001
EU GDPR
Enterprise Cloud
Ask us the hard questions.
Let's schedule a call where you can see the actual product in action, while having your legal, I.T or any other relevant team.
Book a demo