Governance for AI agent actions
Set boundaries for machines.
Keep the choices in human hands.
BlankRealm governs an AI agent’s authority to act. Boundaries come first: identity, purpose, scope, validity, and accountability are checked before every action. Revocation takes effect immediately, and every approval or denial leaves an audit trail.
For product, platform, and governance teams connecting agents to finance, business systems, external APIs, and production data.
The underlying problem
In automated finance, teams often assume an agent will know where to draw the line. Once it can write to real systems, that assumption offers no control. A meaningful boundary must be enforced before a payment is made, with evidence that can be reviewed afterward.
The result: capability is treated as authority, and accountability begins only after something goes wrong.
The result: actions can proceed with boundaries and accountability already in place.
Core capabilities
Use declarative rules to define who can perform which action on which system, and under what conditions.
Validate requests before they reach the target system. Block out-of-scope actions at the boundary, without relying on alerts after the fact.
Record every request, approval, and denial, with the exact rules and timestamps behind each decision.
Policy changes take effect in real time, without code changes or service restarts.
action: payment.create
actor: finance-agent-07
purpose: vendor.settlement
scope:
supplier: supplier_184
amount_lte: 12500 CNY
valid_for: 300s
on_mismatch: deny
decision_record: requiredThe boundary of an action
This is an interactive demonstration. You decide whether the action proceeds. Revoke authorization, and the boundary closes immediately.
Not yet authorized. The action remains behind the boundary.
How it works
Decision Record · Illustrative example
This example illustrates a record structure, not a real customer or transaction. Every decision must identify who acted, why, within which scope, until when, and who is accountable if something goes wrong.
Use cases
Bring risk alerts, case review, field-level evidence access, and audit records into one workflow, with clear permissions and accountability at every step.
Bring internal system writes, ticket workflows, and data exports within explicit authorization boundaries.
Make every order and cancellation explicitly authorized and traceable, even at high decision volumes.
Physical actions cannot be undone. Boundaries must take effect before a command is issued.
Give automated workflows room to operate safely, without constant human supervision.
Why BlankRealm
The world is still taking shape. Any force that tries to define, plan, or optimize it once and for all risks mistaking a provisional answer for a final one.
Machines can execute, verify, track, coordinate, and settle. Their capabilities do not automatically confer authority. Every real-world action must reestablish identity, purpose, scope, risk, and accountability.
Boundaries make machine actions trustworthy. Leaving room keeps time, choices, and possibilities open for people and the future.
BlankRealm turns that restraint into executable rules. Every action must answer a fresh question: is it still authorized, here and now?
Research and perspectives
Where we are today
Available to review today:Our governance model, five-layer decision architecture, interactive prototypes, and research texts.
You can read or interact with all these materials on this site. Where public verification is not possible, we do not ask for trust in advance or present aspirations, plans, or fictional examples as facts.
Tell us which machine action should never depend on assumed trust. We will start there, defining identity, purpose, scope, validity, and accountability.
Your email goes directly to the team. It will not add you to a marketing list.