Previous technological breakthroughs changed what humans could do. AI changes what machines can do on our behalf, including making decisions that previously required human judgement. The challenge is ensuring governance evolves alongside this shift, keeping human judgement, values and authority in control of increasingly autonomous systems.

Founded in 1980 as an information systems engineering consultancy, ADT has evolved alongside applied computer science with a continuing aim: to maximise technology’s benefits through engineering discipline, governance and responsible innovation.

Our vision is a future where intelligence amplifies human capability, enabling wiser decisions, stronger societies and lasting prosperity.

---

A NEW STANDARD FOR AI AGENT MANAGEMENT

AI agents are becoming extraordinarily capable. But capability alone is not enough. An agent that can act at scale must operate within a system that embodies the characteristics of good governance.

Rather than placing governance around AI through policies, prompts or monitoring, ADT builds human authority and systemic control directly into the architecture through external orchestration, mechanical execution control, living standards and coherent multi-agent state. Governance around AI asks it to behave responsibly; governance within the architecture controls what it is permitted to do.

This distinction also makes AI safer for the people and organisations that use it. AI agents do not receive unrestricted access to systems, files or data simply because they have the capability to access them. Their authority is constrained to the task, role and resources they have been explicitly permitted to use. This reduces the risk of unintended changes, data exposure or actions beyond the scope of the user's intent, while allowing the AI to concentrate on the work it has been asked to perform.

The result is a system in which the intelligence can change without changing the governance. Antigravity, Claude or whatever harness or model comes next can be assigned to different roles and tasks, while the specifications, governing standards, authorised system state and evidence of execution remain outside the model and under human authority.

This distinction is fundamental. The AI is not the authority that governs itself. Its actions are constrained by governance mechanisms external to the AI model, and governed actions can be mechanically permitted or denied. ADT can identify standards, regulations and other requirements that appear relevant to the user's intent and present them for consideration. The human remains the authority to determine which requirements and standards are adopted. Once adopted, they become part of the governing specification and are carried through the work as enforceable criteria. Because the system's governing state is held externally, multiple AI agents can work together on complex tasks without any one agent needing to hold the entire system in its own memory.

ADT therefore defines AI agent governance through five essential outcomes: Direction, Performance, Accountability, Transparency and Legitimacy. These are not treated as aspirations or post-hoc checks, but as properties that must be supported by the system's architecture.

Other approaches may provide valuable individual safeguards — and many do. But where a governance approach leaves one of these essential dimensions materially unaddressed, the resulting agent system has a corresponding governance weakness. Safety is only as strong as the governance that constrains capability.

ADT brings these requirements together into a single, inspectable and enforceable model. It separates intelligence from authority, capability from permission and execution from trust.

ADT establishes a new standard for safe, accountable and sovereign AI agent management — enabling AI to become more capable without surrendering human direction, control or authority.

---

THE ADT FRAMEWORK

ADT brings together three complementary dimensions of governance, each addressing a different question: the five outcomes define what effective governance should achieve, the seven stages define how human intent progresses from desired outcome to governed execution and the five pillars provide the architectural mechanisms that make that governance enforceable, traceable and auditable.

Seven stages — the ADT governance model defines a continuous chain from human intent to autonomous delivery. The stages provide the governance structure within which the technical architecture operates, with each stage adding a distinct layer of definition, authority or control.

• Human intent — an operator defines the desired outcome in natural language.
• Specification — ADT translates intent into a governing specification.
• Standards alignment — relevant standards, regulations and requirements are identified and presented for consideration and adoption.
• Decomposition — approved intent is transformed into role-scoped specifications and executable tasks.
• Governed execution — AI agents operate within defined roles, permissions and transfer-control boundaries.
• Traceability — execution remains accountable and auditable, with actions traceable to approved intent and specifications.
• Autonomous delivery — AI agents can design, build and operate systems without requiring humans to write the underlying code, while remaining subject to the specifications, permissions and controls established by the framework.

Five pillars — together forming the Capability Governance Architecture (CGA) — provide the technical mechanisms through which the seven-stage governance model is enforced and evidenced in execution.

• Authoritative Data Source (ADS) — the authoritative, append-only record of the transformation from human intent to governed execution. Significant state changes and agent actions are captured in a SHA-256-chained event ledger, creating an evidence trail through which outcomes can be traced back from execution, through specifications, to the approved intent.
• Specification-Driven Development (SDD) — a strict “no spec, no code” discipline in which every technical change must be covered by an approved specification before execution. Development is therefore governed by explicit, human-approved intent rather than allowing agents to generate or modify code outside an authorised scope.
• Digital Transformation Control Protocol (DTCP) — a privilege-separated enforcement mechanism that validates agent actions against role jurisdiction, specification authorisation and tiered protections in real time. Unauthorised actions are denied, not merely logged, creating an active control boundary between autonomous agents and the systems they are permitted to affect.
• Agent Isolation — separates agents from protected systems through controlled execution environments and restricted resource access, preventing direct modification of protected files, systems or configurations and containing the effects of errors or unintended actions.
• Standards Layer — defines the principles, obligations and governance requirements adopted by the organisation, incorporating relevant external standards and regulations, internal governance policies and documented decisions explaining their adoption, adaptation and rationale.

Together, these mechanisms form a coherent governance architecture in which intent, specification, authority, execution and evidence remain connected. Increasingly autonomous AI can therefore operate within defined boundaries and for purposes established by humans.

Other platforms apply fragmented controls around AI, whereas ADT integrates governance directly into the system architecture. It connects human intent, specification and standards with mechanically enforced execution, rather than relying solely on controls applied around the system.

Its open-source architecture is available for transparent evaluation, collaborative development and wider adoption, while cryptographically traceable execution provides an evidential connection between what was authorised and what occurred.

Taken together, this provides an end-to-end architecture spanning human intent, specification and standards through to mechanically enforced execution and cryptographically traceable outcomes. Based on our review of the emerging AI-agent landscape, ADT appears to occupy a distinctive position through this integration of governance, enforcement and traceability.

---

OPEN AND ACCESSIBLE

ADT makes governed AI and digital transformation freely accessible to individuals and organisations of all sizes. Licensed under AGPL-3.0, it provides the governance foundation, technical architecture and reference implementations needed for transparent evaluation, collaborative development and responsible adoption.

ADT’s latest release establishes an operational framework while inviting community engagement and feedback. Performance and optimisation will continue to evolve as the open-source platform develops.

The open approach is intended to support more than access to the framework itself. Over time, it could provide a foundation for sharing human ideas, intentions, knowledge and governed solutions, with AI helping people discover connections and turn those contributions into practical outcomes while human authority remains central.

Contact Us & Collaborate ➤

FROM GOVERNED AI TO WIDER PARTICIPATION

Embedding governance within execution is fundamental to protecting people, organisations and society from unintended or harmful outcomes. It helps ensure that increasingly autonomous AI systems remain aligned with legitimate human purposes, rather than requiring people and organisations to adapt to AI systems whose objectives may not reflect their interests or the public good.

But the significance of governed AI may extend beyond safety and control. If AI can reduce the technical barriers to creating software, while governance makes that software safe and accountable to deploy, the ability to create digital solutions could become accessible far beyond traditional software organisations. The person who understands a problem — whether a doctor, teacher, lawyer, community organisation or small business — could increasingly become the person able to build the solution.

This could eventually enable something larger: a shared, human-created repository of problems, intentions, knowledge, standards and governed solutions. Rather than simply sharing software, such a system could preserve the reasoning behind it — why a solution was created, what it was intended to achieve, what standards were considered, what choices were made and what was learned from its use.

AI could help people discover connections across this growing body of knowledge and turn human intentions into practical solutions. But authority would remain human: people contribute the ideas, determine what they wish to adopt and decide what should ultimately be built.

In this sense, ADT could become more than a framework for governing AI. It could provide a foundation for communities to share and develop governed digital solutions — by humans, for human purposes, with AI helping to turn ideas into action.

At a larger scale, the same principles could support collaborative systems for challenges that cross organisational and national boundaries: planning for disasters and coordinating mitigation and rescue; environmental intelligence supporting respectful and sustainable ways of living; preserving linguistic and cultural diversity through the translation of music, language and cultural knowledge; and, say, resource ownership and recovery systems in which materials are tracked, reused and valued rather than simply treated as waste.

If continued research and large-scale deployment validate this architectural platform, ADT has the potential to become a foundational enabling technology for Industry 6.0 — an industrial revolution defined not simply by autonomous intelligence, but by autonomous intelligence operating within transparent, accountable and human-governed boundaries.

INDUSTRIAL EVOLUTION

• Industry 1.0 – Mechanisation through water and steam power.
• Industry 2.0 – Electrification and mass production.
• Industry 3.0 – Computing, electronics and automation.
• Industry 4.0 – Connected digital systems, cyber-physical systems and the Industrial Internet of Things.
• Industry 5.0 – Human-centric, sustainable and resilient industry.
• Industry 6.0 – Enables greater capability by embedding governance into autonomous systems.

AI may become a key enabler of Industry 6.0 if humanity determines that its capabilities are necessary to achieve its objectives at scale. The challenge is ensuring that increasingly autonomous systems can be governed in ways that enable greater capability while remaining aligned with human purpose.

Industry 6.0 begins when governance becomes embedded in increasingly autonomous systems, enabling greater capability
while keeping their purpose and operation aligned with human values.