Papers

The formal paper describing OTANIS (Operational Trust and Authority Normative Integrated System), an execution-time governance architecture for agentic AI systems within the wider set of architectural governance models developed here.

The paper defines how authority is established, validated, and evidenced at the point where actions become irreversible. It introduces execution-time admissibility, authority objects, lifecycle constraints, and fail-closed enforcement under real system conditions.

The formal paper describing MGAG, a model for preserving authority across multi-layered, composed systems, complementing the OTANIS framework as part of the same architectural governance programme.

The paper addresses governance across organisational, technical, and regulatory layers, focusing on authority propagation, delegation constraints, refusal pathways, and audit survivability across system boundaries.

The formal paper describing ISDAIRE, which defines the irreducible ex-ante conditions required for governance to exist at all within the architectural governance programme that includes OTANIS and MGAG.

The paper specifies what must be defined before execution: intent, scope, domain separation, authority source, irreversibility awareness, risk framing, and execution boundary. Without these structural definitions, runtime controls cannot substitute for missing governance.

The formal paper describing ARETABA, which defines the minimum execution-time control surface required for governance to operate under real conditions at the boundary where actions become irreversible.

The paper addresses authority, refusal or halt, escalation, traceability, accountability, boundary definition, and admissibility enforcement so that governance remains enforceable, non-bypassable, and fail-closed.

The formal paper describing OTANIS-USG, an extension to the OTANIS architectural governance family for bounded agentic systems with mixed workflow types.

The paper defines how transitional and reversal workflows can support irreversible execution-bearing workflows while preserving boundary discipline, auditability, and fail-closed governance at true commit points.

AI governance is entering a new phase. Many systems no longer only advise, summarise, classify, or generate text—they are increasingly connected to workflows where they may approve, reject, route, trigger, update, grant, revoke, pay, disclose, dispatch, or otherwise change operational state. Once AI can act, control must exist where system behaviour becomes operationally consequential.

Much current AI governance language was developed for advisory systems, model evaluation, and human-led decision workflows. It often lacks precise terms for execution-bearing agentic systems, runtime authority, irreversible operational effects, non-bypass enforcement, fail-closed behaviour, and procurement-grade control evidence. This reference lexicon proposes a structured vocabulary for Architectural Governance (AG) in that context—not to create unnecessary jargon, but to give buyers, engineers, auditors, architects, risk leaders, and governance professionals clearer language for where control must exist, what must be checked, when authority must be valid, and what must happen if execution is not legitimate.

OTANIS is used as the reference architectural family because it already distinguishes ex-ante admissibility, runtime authority construction, execution-boundary enforcement, compositional governance, and multi-layer governance—integrating ISDAIRE, ARETABA, GAG, and MGAG rather than treating governance as one undifferentiated layer. Terms include established engineering and governance language, existing OTANIS-family terms, and proposed AG terms for recurring operational failure patterns that are currently under-described.

Business oriented versions of the OTANIS and MGAG models, written for executives, investors, and operational leaders who need agentic AI governance context without the full formal paper.

They avoid formal notation and focus on clarity, applicability, and decision relevance across architectural governance models at a business readable level.

Plain English explanation of how high-consequence agentic AI systems are typically architected, why conventional control stacks can leave dangerous gaps, and how OTANIS addresses the execution-time governance problem.

It is written for readers who need a practical understanding of why apparently well-governed AI systems may still be dangerous if authority, admissibility, refusal, escalation, traceability, and auditability are not enforced at the moment the action becomes real.

Plain English explanations of the OTANIS framework and MGAG model, designed for accessibility without loss of accuracy in describing execution-time governance ideas.

They are suitable for readers without a technical background who need a clear understanding of architectural AI governance and AI system auditability at a high level.