Ethical AI
Ethical AI, built in.
AI should be useful. And accountable. These are the decisions taken while the products are being written, not a policy page written afterwards: your data never trains a model, people make the decisions, and what runs under the hood is published.
Principles
Six decisions behind every product.
Secure by design
UK-hosted on AWS London, each organisation’s data kept logically separate, Cyber Essentials certified.
Your data stays yours
Customer data is never used to train an underlying AI model.
Your IP stays yours
A partner’s intellectual property remains the partner’s.
Proportional automation
In Pupil Progress AI the AI writes the explanation but never the number. In Attendance Officer AI a trigger is a prompt for professional judgement, not an automated decision. FounderOS never sends anything by itself.
Safeguarding by context
A product used by children carries automated safeguarding monitoring with alerts to your own administrator; a product used by car dealers does not need it, and does not have it.
Controls you administer
Model access is controlled by an administrator, who decides what their organisation can use. Admin audit logs show who accessed what and when.
Show your working
What runs, where, and how the risks are rated.
askKira publishes what is under the hood: which models run, where data is hosted, what is never used for training, how long anything is kept and how the risks are rated. One label for education, one for organisations.
Knowledge owned by the people who hold it
Citadel, and the book behind it.
Citadel comes from the argument of Average, the book by askKira’s founder Mark D Unwin, about AI, institutions and the future of human expertise.
Questions about ethical AI
Ethical AI, answered.
Is our data used to train models?
No. Customer data is never used to train an underlying AI model, and a partner’s intellectual property remains the partner’s.
Where is it hosted?
In the United Kingdom, on AWS London.
What about children?
Products used in schools carry automated safeguarding monitoring, with alerts routed to your own administrator rather than to us.
Does the AI in Pupil Progress AI calculate the numbers?
No. In Pupil Progress AI every figure is calculated from the trust's own data by code, and the AI writes the explanation but never the number.
Does Quote Writer make up prices?
No. If Quote Writer cannot find a reliable price, it leaves the line for you to fill in rather than inventing one.
Does Attendance Officer AI make decisions or send letters automatically?
No. In Attendance Officer AI a trigger is a prompt for professional judgement, not an automated decision, and nothing is sent to families automatically.
Who can see or use my Citadel recordings?
Only you can see what you give Citadel unless you choose to share it. Everything is encrypted and only ever licensed if you tick a box saying it can be.
Does Talkology diagnose children?
No. Talkology does not assess, diagnose or stand in for a speech and language therapist; it offers a hypothesis, not a diagnosis.
Ask us how it works.
Every policy, the DPA and the security answers your data protection officer needs.