Responsible AI
AI with purpose. People at the centre.
Zynaut uses machine reasoning where it helps a person see their own evidence more clearly — never to judge the person. Every system has a stated purpose, a stated boundary, and a person accountable for what it proposes.
HumanPotential.AmplifiedResponsibly.
Where machine reasoning stops
Design rules, not metrics.
- It never scores or ranks a person
- It never advises — a person does
- It shows its reasoning
- It can be contested
- Human-centricAugments a person's judgement, never replaces it
- Fairness under reviewExamined across the groups it touches
- TransparentReasoning in language you can read
- AccountableA named person for every consequential outcome
- BoundedWhere it should not be trusted, stated
Our AI principles
Guiding responsible engineering.
Six commitments about how Zynaut builds systems that reason. They are design commitments — deliberately not compliance claims.
- Human oversightA person can inspect, question and override what a system proposes.
- ExplainabilityReasoning in language the affected party can understand.
- Stated boundariesWhere a system should not be trusted, made visible.
- Data minimisationCollect what the system genuinely requires, for a defined period.
- Fairness under reviewExamined across the groups it touches, and again as it changes.
- Evidence over assertionWhere an outcome has not been measured, we say so.
From design to improvement
Responsibility at every stage.
What is checked before a system that reasons is allowed near a person's record — and what keeps being checked after.
1. Design
Purpose and boundaries. What the system is for, what it must never do, and the conditions under which it should not be trusted — written before anything is built.
2. Develop
Data with provenance and permission. Nothing is used to build a component without knowing where it came from and under what consent it may be used.
3. Test
Behaviour across the groups it touches. Where a system affects opportunity, its behaviour is examined across those groups, and the examination is repeated as its inputs change.
4. Deploy
A human accountable. A named person can inspect, question and override what the system proposes, and owns any consequential outcome.
5. Improve
Corrections returned only where permission allows. What was wrong goes back into the foundation under a declared purpose and a lawful basis — never as a by-product of use.
Built on trust
Responsible AI in action.
Every reasoning component in Pathways is built inside a stated boundary, with the provenance of its inputs known, its behaviour examined, and a person accountable for what it proposes. What it produces is reasoning a reader can follow — never a verdict about a person.
- Purpose declared before a line is built
- Data used only with provenance and permission
- Behaviour examined across the groups it touches
- A person can inspect, question and override
- Reasoning shown in language the affected person can read
Technology is powerful when it serves people fairly.
- Privacy
- Consent and purpose
- Security
- Access and permission
- Provenance
- Human oversight
- Audit
- Correction
Where AI sits in Pathways
One component of a system, never the system.
Four rules about the place machine reasoning holds in the record — and the places it is refused.
It never scores or ranks the person
Confidence is a fact about the evidence, claim by claim. It is never added up into a number about you and never compared between people.
- No score, no rating, no league table
- No comparison between people
- Confidence describes the evidence
It never advises — a person does
The record shows the gap and the reasoning. A counsellor, mentor or adviser gives the advice, and their advice is recorded beside it.
- Guidance comes from a person
- The reasoning is shown to both of you
- No chatbot stands in for a counsellor
It shows its reasoning
Every conclusion lists the evidence it used and the evidence it left out, with a reason for each. A conclusion that cannot show that is not one the record makes.
- What was counted, and what was not
- Why, in plain words
- What would have to change
It can be contested
Every conclusion carries a route to challenge it, and a person reads the challenge. The system does not judge its own appeal.
- A contest route on every conclusion
- Read by a person, not the system
- Recorded beside the conclusion
- Human oversightA person accountable for consequential outcomes
- Explainable by designReasoning a person can read
- Stated boundariesWhere it should not be trusted, published
- Data minimisationCollect what the system genuinely requires
- Fairness under reviewExamined, and examined again, across groups
- Evidence over assertionUnmeasured outcomes are called unmeasured
Questions people ask
Responsible AI, in plain answers
Does AI decide anything about me?
No. Machine reasoning in Pathways helps organise evidence and show a gap. A person makes every consequential decision — an adviser, an institution, an employer, or you — and what the system proposes can be inspected, questioned and overridden by that person.
Is the AI free of bias?
No system can promise that, and this site does not. What is committed to is review: where a system affects opportunity, its behaviour is examined across the groups it touches, and the examination is repeated as the system and its inputs change.
Is my record used to train Zynaut's models?
Not by default, and not as a by-product of use. Nothing leaves an offering without a declared purpose, a lawful basis and a permission covering that use. Corrections return to the foundation only where purpose, permission and policy allow.
Can I see why the record reached a conclusion?
Yes. Every conclusion lists the evidence it used and the evidence it did not count, with a reason for each, in language written for the person it is about. A conclusion that cannot show that is not one the record makes.
Are there published figures on oversight, bias or audits?
No. No independent audit has been completed, and no incident count is measured over a period a reader could check, so none is printed. When such a figure exists it will be published with its method and its dates, not as a headline.
Who do I speak to if a conclusion seems unfair?
A person. Every conclusion carries a route to contest it, and a person reads the challenge; the system does not judge its own appeal. The contest is recorded beside the conclusion, so anyone who later sees one sees the other.
