AetheriaIntelligence

Advisory · Technology · Commercial · Research

Advance with AI.
Stay in command.

AI can take on more of the work. Your organization still has to decide what it is allowed to do—and answer for what follows.

We advise on those decisions and develop the technology to carry them into day-to-day operations.

AI governance, from policy to practice.

From the Aetheria textbook

For people responsible for the decisions AI helps make.

Decision rightsOperational controlsEvidence

Our approach

Delegation should
keep its limits.

Giving an AI system a task leaves it with choices: which records to use, what to recommend, perhaps whom to contact. Those choices can exceed what the person assigning the work intended.

We focus on keeping the original authority intact as a request becomes a specific action. That starts with the organization’s own decisions about access, discretion, and responsibility.

Some work can proceed under a standing mandate. Other decisions need a person’s judgment. The controls should reflect that difference.

Work through an AI use case
From the Aetheria textbookInformation and action
Handwritten textbook diagram showing a research answer and a separate protected action, with authorization, execution, and observation steps.
Reading a record, sharing an answer, and changing a record each raise their own permission question. View full illustration ↗

A municipal example

Fifteen years of records.
One question from a clerk.

A clerk wants to know how the council has handled a recurring issue. An AI assistant could help trace the discussion across years of minutes, provided the clerk can check the sources and understand where the record is incomplete.

The next request may seem small: mark a record as reviewed. But that changes stored information. It needs permission of its own, tied to the record the clerk actually examined.

See the proposed evaluation
Council recordsIllustrative workflow

What would the clerk need to check?

  1. Research

    The records behind the answer

    Search follows the clerk’s permissions. Citations let the clerk return to the source; coverage limits explain what the answer can support.

  2. Review

    The proposed change

    An internal marker changes from unreviewed to reviewed. The authoritative minutes are left untouched.

  3. Confirmation

    What was actually recorded

    The change is authorized for the expected record version. A separate readback checks the result; missing confirmation leaves it unresolved.

Proposed evaluation scenario. This example describes no deployed customer system.

Aetheria: Delegated Authority and Execution Governance, by Bradley D. Wilson. Navy architectural gateways with a gold path.

2026 textbook manuscript

Research and education

Authority is also
a question about people.

A system can check whether a rule was followed. Whether that rule gives the right person the right discretion is a further question.

Our research agenda examines both: how delegation can be represented in software, and what makes it justified and open to challenge. The textbook develops these questions through cases, diagrams, and exercises.

Explore the research agenda

Working manuscripts and proposed studies; university collaborators welcome.

Get in touch

Where is your organization considering AI?

Tell us about the work, the people responsible, and the decision you are trying to make.

Contact Aetheria