Constitution before policy
Immutable protections remain distinct from organization policy, so a local rule cannot quietly lower a foundational boundary.
Operational conscience & authorization layer
Intelligence proposes.Conscience evaluates.Authority is established.Only then may action occur.
EthosGate is runtime infrastructure for deciding whether a consequential AI action is authorized to happen. It is not an ethics dashboard, chatbot, or decorative score.
Initial release scope: shadow and advisory modes
Decision request
Shadow evaluation
Action
release_customer_recordContext
insufficient evidenceConsequence
materialAuthority
review requiredProposed decision
ALLOW
MODIFY
QUESTION
ESCALATE
REFUSE
ResultQUESTION · evidence needed
Conceptual interface · no enforcement claim
Action lifecycle
Every consequential action needs a visible path to authority.
EthosGate organizes the decision boundary from proposal to record. At every step, the system is designed to preserve what is known, what is uncertain, and who is accountable.
01
Intent
An agent proposes a consequential action.
02
Context & evidence
The request, provenance, permissions, and uncertainty are preserved.
03
Consequence
Potential impact is classified before a decision is considered.
04
Invariants & policy
Non-negotiable protections take precedence over organization rules.
05
Decision
The action is allowed, changed, paused, escalated, or refused.
06
Human authority
Authorized people decide when a consequential boundary requires judgment.
07
Authorization
Only an exact, reviewed action can receive a bound authorization.
08
Protected execution
A future enforcement boundary verifies authorization before action.
09
Verifiable record
The decision, authority, and outcome remain inspectable.
One vocabulary, everywhere
A decision should be readable at a glance—and traceable in detail.
The same outcome names carry through the Workbench, review workflow, audit record, and API response.
ALLOW
The evaluated action may proceed within its stated limits.
Decision stateMODIFY
The proposal needs defined changes before another evaluation.
Decision stateQUESTION
Evidence or context is incomplete, conflicted, or missing.
Decision stateESCALATE
The action pauses for authorized human review.
Decision stateREFUSE
The action is blocked by a protection or unacceptable risk.
Decision stateDesigned for consequence
A governance boundary, not a confidence theater.
The architecture gives evidence, constitutional protections, human authority, and decision provenance a first-class role before a high-impact action becomes real.
Evidence that can be inspected
A decision is accompanied by its sources, provenance, uncertainty, and relevant context—not a black-box assurance.
Human authority where it matters
High-consequence proposals are designed to pause for an accountable human decision rather than manufacture confidence.
Authorization bound to the action
The architecture is designed to bind authorization to a specific action, actor, target, parameters, and expiry.
Records made for verification
Decision lineage, rationale, authority, and observed outcomes are structured for a tamper-evident audit trail.
Initial operating modes
Learn the boundary before you depend on it.
EthosGate is positioned for controlled evaluation first. Shadow and advisory use make decisions visible without representing that a live action is being technically enforced.
Shadow mode
See the decision without changing execution.
Compare proposed actions with EthosGate’s evaluation path and build an evidence trail while existing operations remain unchanged.
Advisory mode
Bring human reviewers into the decision.
Surface evidence, uncertainty, and required authority for accountable review—without claiming production enforcement.
Conscience. Authority. Action.