UNA

AI should help you without taking control.

UNA is a way of building AI that shows what it knows, admits what it does not know, and stops when a person needs to decide.

In plain English: You stay in charge. The AI must follow clear limits and leave a record of important actions.

View the UNA system mapWhat this means for you
Research in progress · tested locally · not a public product
Interactive UNA system map · public overview

See where assistance ends and human authority begins.

This browser-local SVG explains UNA’s intended governed path. Select a node to inspect it. It is an informational map—not live telemetry, permission to act, or evidence that an external action occurred.

Interactive UNA governed workflow system map Select any of seven steps to inspect its input, output, and authority boundary. Arrow keys move between steps. 01 · Human intentAsk a question andchoose the context 02 · KnowledgeRetrieve reviewedknowledge 03 · GuardrailsCheck uncertainty,risk, and authority 04 · ExplanationOffer a bounded,reviewable option 05 · Human authorityThe authorized persondecides what is next 06 · RecordPrepare an authorizedlocal proposal 07 · BoundaryExternal executionremains separate
Choose a node or use the arrow keys. Suggestion, selection, approval, execution, publication, and memory remain distinct states.
  1. Human supplies a question and chosen context.
  2. UNA retrieves reviewed knowledge.
  3. UNA checks uncertainty, risk, and authority.
  4. UNA offers a bounded explanation.
  5. The authorized person decides what happens next.
  6. UNA can prepare an authorized local proposal.
  7. External execution remains separate.
Inspect the Mermaid source
flowchart LR
  HUMAN["Human supplies a question and chosen context"]
  KNOW["UNA retrieves reviewed knowledge"]
  CHECK["UNA checks uncertainty risk and authority"]
  EXPLAIN["UNA offers a bounded explanation"]
  DECIDE["The authorized person decides what happens next"]
  RECORD["UNA can prepare an authorized local proposal"]
  BOUNDARY["External execution remains separate"]
  HUMAN --> KNOW --> CHECK --> EXPLAIN --> DECIDE --> RECORD --> BOUNDARY
Why this matters

What could this change in everyday life?

UNA is being designed so AI can be useful without quietly gaining more power over your work, information, or choices.

At work

You could see why the AI made a suggestion, what information it used, and when a person must approve the next step.

With your information

The system would use clear limits. Learning more about you would not automatically give it more control or permission.

When something goes wrong

You could trace what happened, challenge the result, correct the record, and reverse an action when reversal is possible.

A simple way to think about it

UNA is the rules, memory, and receipt book around AI.

The AI may help with thinking or a task. UNA is meant to keep that help inside limits people can understand and control.

The rules

What the AI may do, what it may not do, and when it must stop.

The memory

What should be remembered, where it came from, and who may use it.

The receipt

What happened, why it happened, who approved it, and what changed.

The design rules

Six things UNA is being built to do.

These are design goals. Some parts exist and have local tests. The whole system has not yet completed independent testing.

01

You stay in charge

The system should support your judgment, not replace it or pressure you to agree.

02

It shows its work

You should be able to see the evidence and rules behind an important result.

03

It says when it is unsure

The system should make doubt visible instead of hiding it behind a confident answer.

04

Important actions pause

High-impact steps should stop for the right person to review and approve them.

05

Memory has limits

The system should learn what is useful without collecting more data or power than it needs.

06

It leaves a receipt

Important actions should leave a clear record that people can inspect and challenge.

How it works

Five steps from your request to a clear record.

01

You ask

The system receives your request and the context you chose to share.

02

It checks

It looks at the evidence, uncertainty, risks, and rules that apply.

03

It suggests

It offers a result or a possible next step within those limits.

04

You decide

If the step needs human approval, the right person must approve it.

05

It records

It saves what happened, why, who approved it, and what changed.

A receipt for important AI actions

One record. Five questions.

This is meant to help people reconstruct what happened without accepting one system-made story.

AIWhich AI was used?
SituationWhat did it know at the time?
PersonWho made the decision?
PermissionWhat was it allowed to do?
ResultWhat actually changed?
For technical readers: system-level causal record

UNA research separates model identity, environment state, human responsibility, authority, evidence, and external effects while keeping them linked. The goal is record integrity and reconstruction. A complete record would not by itself prove that one causal explanation is true, that the system is legally compliant, or that the full system has been independently validated.

Where the work stands

UNA is research in progress, not a finished public product.

Today
Active research and development
Testing
Local tests exist for some parts
Whole system
Integration and independent testing are incomplete
Public use
Not deployed for public operation
Learn more

Useful AI should be easy to question.

Explore the public work and evidence behind UNA, or contact me to discuss where governed AI could help people in practical ways.

Explore the workSee the evidenceContact Tom