Trust, governance & security

More AI capability should not mean less human control.

EAOS is designed around explicit authority, least privilege, evidence, rollback and progressive qualification so intelligent systems can help without silently acquiring consequential power.

Human Authority

Consequential actions are separated from advice and preparation. Approval remains a real operational gate.

Least Privilege

Identity-first access, scoped permissions and role separation are architectural requirements.

Evidence First

Material actions should record who, what, why, before/after state, outcome and rollback path.

Private by Design

Local, sovereign and provider-neutral deployment patterns reduce unnecessary dependence and exposure.

Sandbox First

Dry run, staged promotion, acceptance checks and rollback separate candidates from trusted operation.

Claims Discipline

Live, demonstrated, architecture and frontier states are visibly separated rather than collapsed into marketing language.

Authority chain

Advice -> preparation -> approval -> execution -> evidence.

The exact mechanics differ by service, but the control principle remains consistent across AI workflows and hosting operations.

1

AI Advice

Explain, diagnose, analyse or recommend.

2

Change Plan

Prepare the intended action, impact and rollback.

3

Policy + Approval

Check identity, permission, scope and authority.

4

Controlled Execution

Run only through an approved pathway.

5

Verification + Evidence

Confirm the result and preserve the record.

Public website rule

Never turn a design approval into a production claim.

Live / Verified: specific current evidence supports the claim.

Demonstrated: a working foundation has been exercised but may have productisation limits.

Approved Architecture: the design is authorised but still requires implementation and qualification.

Frontier / Vision: a long-range research target, never represented as currently live.