Framework / Governance Model
Governance & Trust Architecture Framework
GTAF is TNT's governance model for delegated action in AI and automated systems. It addresses a practical gap many organizations now face: policy says oversight is required, but operating models for legitimate delegation are often still missing.
Instead of treating governance as prose, GTAF treats boundaries, decision layers, responsibility, and validity as explicit artifacts. That makes the model usable for real implementation, not just for conceptual positioning.
Public normative reference
A governance framework for AI systems that may act, delegate, or produce consequential effects. GTAF turns scope, authority, responsibility, and validity into structured operational artifacts.
GTAF is TNT Intelligence's governance framework for AI systems that may act, delegate, or produce consequential external effects.
Its central contribution is simple but materially different from most AI governance talk: it turns scope, authority, responsibility, validity, and intervention into explicit operational artifacts instead of leaving them distributed across prose, implicit conventions, or after-the-fact review.
Defines what is inside scope, what is outside, and under which conditions a delegated capability is even relevant.
Describes which action may be delegated, with which constraints, conditions, and expected effect.
Makes outcome ownership and escalation responsibility explicit instead of assuming that authority somehow "comes with the system."
States whether delegation is actually permitted in the current scope, version, and validity window.
GTAF can be read as an operationalization layer for many of the practical questions organizations face under the EU AI Act.
It is not legal advice and it does not replace legal interpretation. What it does provide is an operating structure for turning governance obligations into explicit system design and decision artifacts.
| AI Act concern | How GTAF helps operationalize it |
|---|---|
| Risk classification and intended use | GTAF forces explicit scoping of use case, action surface, impact boundary, and validity conditions. |
| Governance and oversight | Responsibility bindings, decision layers, and intervention paths make authority and escalation legible. |
| Documentation and traceability | Structured artifacts and references create a stronger base for technical documentation and auditability. |
| Lifecycle control | Validity windows, revalidation logic, and controlled change keep delegation from becoming permanent by accident. |
GTAF is easiest to understand as a chain that must stay intact from scope definition to permission state:
Delegation is not a property of the model. It is a property of an explicitly bounded and currently valid operating context.
A simplified artifact chain might look like this:
system_boundary:
id: sb.invoice.review.eu
scope: "invoice review up to EUR 5000"
decision_record:
id: dr.invoice.auto-approve
action: "invoice.approve"
constraints:
supplier_status: "approved"
max_amount_eur: 5000
responsibility_binding:
owner: "finance.operations"
escalation: "head-of-finance"
delegation_readiness_check:
status: PERMITTED
valid_until: "2026-12-31T23:59:59Z"This is the key move: GTAF does not say "the agent is trustworthy." It says that a specific delegated action is or is not permitted within a defined boundary and time window.
GTAF is relevant for organizations that need more than AI policy language. Typical collaboration paths include:
structuring AI use cases into decision layers, scope boundaries, and accountable operating models
translating oversight and governance requirements into explicit technical artifacts
reviewing whether a planned AI system can be delegated safely at all, and under which conditions
preparing the path from governance design into runtime enforcement and implementation
When organizations already have policy language but still lack an operational way to define who may delegate what, under which conditions, and for how long.
When AI systems move into approval paths, operational tooling, internal support, or business workflows where scope and accountability cannot stay implicit.
Deterministic runtime enforcement core
The enforcement core that turns evaluated governance outputs into executable allow/deny decisions. The public implementation demonstrates the contract, but the runtime model is broader than one language.
See enforcementIntegration layer around the runtime core
The adoption layer that helps real systems load artifacts, shape execution context, and call the runtime cleanly. The public implementation is one concrete path, but the integration model is not language-specific.
Understand integrationInfrastructure for Trusted Autonomy
A runtime architecture for systems that must act with real effect while staying governable. ITA extends GTAF into execution spaces, capability exposure, enforcement, and audit.
See the architectureFrom GTAF through Runtime and SDK to ITA, TNT already brings public reference work, runtime building blocks, and applied architecture into these questions. The conversation does not have to start at theory.
When these questions move from interest to implementation, TNT is a serious conversation partner.
Discuss your context