Skip to main content
Skip to main content
Transparency

Trust and transparency: how the assistants actually work

We sell governance, so we should be able to evidence our own. This page sets out how a suggestion is produced, what gets recorded, where a human stays in charge, what the assistants cannot do, and which of those claims you can check for yourself today.

Maintained by Andy Wake, Founder and Managing Director · Last reviewed 28 August 2026

Our commitments

You are always told when you are talking to AI

Every assistant surface — chat, avatar video and study tools — carries a visible AI disclosure before you interact with it. The assistants are presented as assistants: they do not claim to be human, a teacher, a clinician or a lawyer, and they say so when asked.

A person stays responsible for the decision

The assistants produce drafts, options, revision plans and challenge questions. They do not make decisions about a person on their own. Anything consequential — a safeguarding concern, a board decision, a grade judgement — goes to a named human who can accept, change or reject it.

Recommendations and consent changes are logged, not lost

Governance events — a recommendation, a nudge, a wellbeing check-in, a consent change, a data export, an override — are written to an append-only record with the reason attached. That is what lets you reconstruct why the assistant said what it said, rather than taking our word for it.

Memory is opt-in and reversible

Personalisation depends on remembering context, so memory is tied to consent you give and can withdraw. Account data is scoped to the signed-in person by database access rules, and you can ask for a copy or deletion of your data at any time.

Accessibility is tested and the results are published

Accessibility testing runs against the site continuously and the findings are published — including the history of previous audits, so you can see what improved and what has not. Avatar videos ship with captions and transcripts, and Sign Languages provides BSL-informed support.

The rules we work to are written down and public

Privacy, cookies, security, ethical AI, accessibility, intellectual property, payment compliance, complaints, refunds and continuous improvement are all published policies rather than internal notes. If a policy and the product disagree, the policy is the commitment and we fix the product.

What the assistants cannot do

Limits stated plainly are worth more than reassurance. These are the boundaries we hold ourselves to.

  • The assistants can be wrong. They generate language from patterns, so they can state something plausible and incorrect. Check anything that matters — a grade boundary, a legal duty, a clinical question — against the source we cite.
  • EDIE offers wellbeing support and companionship. She is not a therapist, a counsellor or a crisis service, and she signposts to human help rather than replacing it.
  • SAMU and TRAC support GCSE study. They do not predict grades, mark official work or replace a teacher's professional judgement.
  • We do not publish client case studies, testimonials or ratings yet, because we will not put a claim on this site that a reader cannot verify. When we have consented references, they will appear with the client named.

Raise something with us

Automated Governance Systems Ltd is registered with the Information Commissioner's Office under reference ZB953617.

More about the company and who is accountable is on the about page.

Try the demo