Better decisions

What does "honest AI" mean for business decisions?

Honest AI as an engineering behaviour: a validation gate that grades every answer pass, warn or block.

· 17 Nov 2026 · 2 min read

Honest AI describes an engineering behaviour: the system will not present a number it cannot support. In Unloq’s usage, it means every figure is computed in code from your data. Every answer passes a validation gate that checks each figure against the query that produced it, confirms the definitions behind it are consistent, and grades the result pass, warn or block before a person sees it.

How does the validation gate work?

Every figure goes through the validation gate: pass and warn reach the reader, block is stopped

The gate sits between the answer and the reader. It checks that every figure traces back to the query and the data that produced it, and that the definitions underneath the figure are the same ones used everywhere else in the organisation.

A pass sends the answer through with its evidence attached. A block stops it.

In the demos we run, “warn” is the grade nobody asks about, and it is the one doing most of the work. It covers the answer that is supportable but has something underneath it worth knowing about: thin data, a definition that resolved through an unusual route, a figure sitting at the edge of what its source can carry. A leader can act on a warned answer. They act on it knowing what they are standing on.

Everything the gate does is recorded, so a recommendation can be traced back to its source months later by somebody who was not in the room at the time.

What goes wrong when an AI answer sounds sure regardless of the data?

Two confident answers: one with its query, data and definitions, one with none shown

The default failure mode of a general-purpose AI assistant is fluent confidence: an answer delivered in the same tone whether the data supports it or not. In the first draft of an email that costs little. When the output feeds a decision with budget attached, the reader has nothing in the tone to tell a supported answer from an unsupported one.

Is “honest AI” a recognised standard?

The phrase is our own shorthand. The principle underneath it is well established: the US National Institute of Standards and Technology’s AI Risk Management Framework, published in January 2023, lists “valid and reliable” first among the characteristics of trustworthy AI and calls it a necessary condition of trustworthiness, the base the other characteristics rest on.

What honest AI does not fix

The gate checks the answer against the data; the question against the business is yours

The gate checks the answer against the data. It cannot check the question against the business. If an organisation has three working definitions of “active customer”, a system that resolves them consistently will return one number, reliably, indefinitely, and the argument about which definition should have won is still yours to have. We have not solved that one, and we are wary of anyone who says they have.

Related questions- What is governed decision intelligence?- What is an Impact Ledger?

Written by

The Unloq team

Unloq is a decision intelligence company in Birmingham. We turn the data a business already holds into recommendations a leadership team can question before it commits. Company news, research and the articles we write together are published under the Unloq name.

Got a question about this?

Send it to us and one of the team will come back to you with an answer.

Ask a question

Bring us one decision that matters

Send us the call you have to make. A founder reads it and comes back on it, and if answering it properly needs your own data we will tell you exactly what that would take.

Unloq

Outthink uncertainty. Decision intelligence for your whole business, with the evidence attached.

© 2026 Unloq Solutions Ltd. All rights reserved. Registered in England and Wales, company number 15958795.

Cyber Essentials Plus
Social Icon
Social Icon

Bring us one decision that matters

Send us the call you have to make. A founder reads it and comes back on it, and if answering it properly needs your own data we will tell you exactly what that would take.

Unloq

Outthink uncertainty. Decision intelligence for your whole business, with the evidence attached.

© 2026 Unloq Solutions Ltd. All rights reserved. Registered in England and Wales, company number 15958795.

Cyber Essentials Plus
Social Icon
Social Icon