FUURAA AI Civilization Architecture · Operating-layer engineering dossier
Permissions: make every capability bounded and revocable
Connecting a tool does not grant an Agent blanket permission to use it. Permission decisions should be resource- and action-specific, deny by default, short-lived, re-evaluated at consequential boundaries and independently revocable.
Public statusResearch direction · architecture referenceEvidence statusFUURAA method synthesis grounded in primary standardsSources checked19 August 2026
Core decision
Which exact action, resource, fields, purpose and time window are allowed now?
Use this sequence for architecture review, threat modelling and test planning. It is not a universal compliance checklist and cannot replace system-specific engineering validation.
01
Describe the capability
Name action, resource, fields, amount, geography and explicit exclusions.
02
Bind purpose and principal
Join the capability to one principal, approved purpose and operating context.
03
Evaluate current policy
Use exact policy version, risk class, approvals, conditions and denial reason.
04
Limit time and consequence
Issue short-lived authority with budgets, rate limits and step-up approval.
05
Re-check and revoke
Re-evaluate before consequential calls and prove revocation reaches active workflows.
FUURAA analysisLeast privilege is not a static role label. For Agent systems it is a continuously evaluated relationship among principal, purpose, capability, resource, consequence and time. A technically valid token can still be invalid for the present action; runtime policy must be able to say no after planning has begun.
Minimum interface contracts
Put consequential semantics in inspectable interfaces instead of relying on assumptions between systems.
Field names are public engineering references, not a normative protocol. Implementations may use other structures, but should expose every lost, defaulted or downgraded semantic.
01
Capability contract
Must be intelligible to sender, receiver and independent reviewer.
Minimum fields
verb, resource and field-level scope
amount, frequency and consequence limits
denied actions and environments
Evidence gate
No wildcard capability reaches a consequential tool without an explicit exception decision.
Stop condition
If the contract is unresolved, expired or silently downgraded, block consequential action and route to review.
02
Policy-decision contract
Must be intelligible to sender, receiver and independent reviewer.
Minimum fields
principal, purpose and context
policy version, inputs and decision
approver, conditions and reason
Evidence gate
A reviewer can reproduce which policy evaluated which facts.
Stop condition
If the contract is unresolved, expired or silently downgraded, block consequential action and route to review.
03
Runtime authority contract
Must be intelligible to sender, receiver and independent reviewer.
Minimum fields
issued, effective and expiry time
revocation handle and status
checkpoint and re-authorisation boundary
Evidence gate
Revocation and expiry stop queued, retried and long-running actions.
Stop condition
If the contract is unresolved, expired or silently downgraded, block consequential action and route to review.
Failures to seek deliberately
Verify that boundaries really deny, stop and preserve evidence.
Nominal success cannot establish an effective boundary. Tests should manipulate identity, time, version, network, policy and partial failure while retaining raw outcomes.
T1
Scope expansion
Request a neighbouring field or resource and require explicit denial.
T2
Policy staleness
Change role or policy mid-task and verify re-evaluation before the next effect.
T3
Revocation race
Revoke while calls are queued and prove no later call inherits cached authority.
T4
Approval reuse
Attempt to reuse an approval for another amount, recipient, tool or purpose.
Minimum engineering evidence package
Let the next owner reproduce the decision, open artefacts and see remaining unknowns.
A complete package only makes evidence relationships reviewable; it does not prove artefacts authentic, controls effective, the system safe or the decision correct.
01
Capability inventory
Tool, action, resource, limits, denied operations and owner.
02
Policy decision logs
Versioned inputs, outcome, reasons, approvals and exceptions.
03
Revocation evidence
Propagation time and outcome across tokens, queues, workers and retries.
04
Negative authorisation tests
Wrong scope, audience, purpose, expiry, amount and policy context.
05
Exception review
Who accepted which residual risk, until when and with which reopen trigger.
Applicability boundary
This reference does not define contractual, employment, fiduciary, financial, medical or legal authority. OAuth and zero-trust guidance cannot establish whether a real-world action is lawful or appropriate. No FUURAA product permission capability is claimed.
Primary sources and evidence boundaries
Use standards language without presenting citations as implementation evidence.
Every source states publication timing, its role in this dossier and its non-transfer boundary; living source pages were checked 19 August 2026.
