AI Ethics & Transparency

We build AI systems for organisations that operate under real commercial and regulatory conditions. That constrains what we build and how. These are the principles we apply, and they are commitments rather than aspirations: they show up as clauses in engagement documents and as tests in the systems themselves.

1. A person remains accountable

Every system we build has a named owner on the client side who is accountable for its behaviour. Automation moves work; it does not move responsibility. Where a system acts without a person in the loop, the boundary of what it may decide alone is written down and agreed before it is built.

2. Decision boundaries are explicit

We define, in writing, what the system decides, what it recommends, and what it must escalate. A system that is allowed to act inside a narrow, well-understood band and hand everything else to a person is more useful, and far safer, than one whose limits nobody has written down.

3. Traceability by default

A decision that cannot be reconstructed cannot be defended to a regulator, a customer or a court. Systems we build record what they were asked, which model and version answered, what context was retrieved, what they returned and what happened next. That record is designed in at the start, not added after an incident.

4. Evaluated before release, monitored after

We agree the criteria a system is measured against during the consulting phase, before anything is built. A system ships when it meets them, not when it demos well. After release it is monitored against the same criteria and re-verified as models, data and usage change, because a system that was correct in March can drift by September.

5. Data minimisation and provenance

We use the least data that will do the job, and we ask where data came from and what it may lawfully be used for before it goes into a system. Training or fine-tuning on client data happens only where the client has the rights and has agreed it in writing, and never to improve a model used for another client.

6. Bias and fairness

Where a system affects people, deciding who gets seen, served, priced, hired or approved, we test for disparate behaviour across the groups the client is required or chooses to protect, agree thresholds before launch, and re-test on a schedule. Where we cannot test fairness meaningfully, we say so rather than implying it has been handled.

7. Disclosure

People interacting with a system we build should be able to tell that they are dealing with software. We recommend clear disclosure in every product we help design, and a route to a human that actually works. We will not build an interface whose purpose is to make a machine pass as a person.

8. Robustness and failure

Systems are designed to fail visibly rather than quietly. Confidence thresholds, fallbacks, rate limits, guardrails and circuit breakers are part of the architecture. Where a model is unavailable or uncertain, the system degrades to a defined, safe behaviour instead of guessing.

9. Security and privacy

Ethics work is undone by weak engineering. Our Security Statement and Data Processing Addendum describe how access, data and incidents are handled.

10. Work we decline

We do not build systems whose purpose is to deceive people about who or what they are dealing with, to conduct surveillance of individuals without a lawful basis, to score or profile people in ways prohibited by applicable law, to generate content designed to mislead about real people or events, or to make consequential decisions about individuals with no route to human review.

We will also say when the answer is that a use case should not be automated. That is part of what consulting is for.

11. Transparency about our own limits

We are an engineering firm, not an assurance body. We do not certify systems, our own or anyone else's. Where an engagement needs independent audit or conformity assessment, for example under a sectoral regulator or the EU AI Act, we build so that the assessment is possible, and we say plainly that the assessment itself is somebody else's job.

12. Contact

Questions, or a concern about a system we built: contact@globalconsultai.com. We answer these ourselves.