Skip to main content
TomorrowX

Perspectives

Discover the purpose, scope and potential of Data Mediation.

Start with the Forbes Australia feature, then explore how Data Mediation creates customer-controlled AI enforcement, immediate cyber prevention and repeatable enterprise change without disrupting the systems already in operation.

Kostas Siourthas, founder and CEO of TomorrowX

Start here · Shared introduction

From Legacy to AI

The Forbes Australia feature introduces the enterprise problem, the company history and the role of Data Mediation in connecting AI with the systems that continue to underpin enterprise and public-sector operations. Every research path begins here.

Read the Forbes feature

Shared introduction · The category

Workflow AI

The definition that every other perspective builds on — and why most of the value of AI sits inside the workflow. Workflow AI is the use of probabilistic intelligence upon private company and public-sector agency data, towards delivering a specific product, service, experience or benefit to a specific user, over time.

Read the Workflow AI perspective

International guidance · Last reviewed 22 October 2026

Different authorities. A common question: who remains in control when AI can act?

Across jurisdictions, guidance is becoming more specific about access, identity, approved actions, monitoring and accountability. The details differ, but the practical question is consistent: can the organisation define the boundaries of AI use and demonstrate that they hold in operation? Each entry below links to the primary source and states what kind of document it is.

Australia with CISA, NSA, Canada, NZ and UK · 1 May 2026 · Joint guidanceCareful adoption of agentic AI servicesAdopt incrementally; bind tools and identities to minimum scope; defence in depth; human oversight; monitor and test.United Kingdom · NCSC · 15 May 2026 · BlogThinking carefully before adopting agentic AIStart with low-risk tasks, apply established security fundamentals and prepare for unexpected behaviour.United Kingdom · NCSC · 4 August 2026 · StatementNCSC statement on incidents from frontier AI evaluationsSafeguards, oversight and response plans from the start; cyber fundamentals still apply.Australia · ASD · 11 September 2026 · AdvisoryAgentic AI harnessesThe harness connecting models to data and tools needs least privilege, governance, monitoring and human oversight; a model-only safety strategy is insufficient.Australia · ASD · 29 September 2026 · AlertProtect your organisations’ AI servicesSecure identities and credentials, limit privileges, enforce consumption safeguards, monitor activity and protect connected downstream systems.Singapore · MAS · 7 October 2026 · Guidelines, effective 7 October 2027Guidelines on AI Risk Management for Financial InstitutionsBoards and firms retain accountability, including for third-party AI; proportionate life-cycle controls and compensating controls where needed.European Union · Regulation (EU) 2024/1689 · Applied since 2 August 2026 · LegislationEU Artificial Intelligence ActBinding, risk-based obligations for providers and deployers in scope. Since the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force 27 July 2026), high-risk obligations apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I).International · ISO/IEC · 14 October 2026 · International StandardISO/IEC 27090:2026Understand and address threats and compromises specific to AI systems across the life cycle. Not a certification of any product or deployment.India · CERT-In · 15 October 2026 · Advisory CIAD-2026-0049AI agents as attack vectorsDefault-deny egress enforced at the network layer for evaluation tasks; per-agent identity, task scope, real-time logs and safeguards for enterprise deployments.United States · NIST · 17 February 2026 · Standards initiativeAI Agent Standards InitiativePromotes interoperable and secure agents through industry-led standards and research; not a regulation or certification.United States · NIST NCCoE · 5 February 2026 · Concept paperIdentity and authority of software agentsAgent identities, authorisation and auditing are active areas of standards work; not an adopted mandatory standard.United States · NIST · 27 August 2026 · BlogWhy agentic AI needs a strong identity foundationIdentity and authorisation foundations need to evolve beyond model-only guardrails. NIST commentary, not regulation.

These documents differ in legal status: binding legislation, an international standard, supervisory guidelines with a future effective date, official cyber advisories, a joint best-practice publication and standards initiatives. None of them endorses TomorrowX or mandates a particular technical mechanism. TomorrowX’s own reading of them is set out in The Layer Above Every Harness (a TomorrowX perspective) and ISO/IEC 27090: putting AI security guidance into operation (a standards interpretation).

Choose a path

Explore the part of the work most relevant to each reader.

Each path has a defined sequence. Begin at step one and move through the perspectives in order, or go straight to the one that matters most. Every path finishes at the platform, an applied solution or the next stage of the work.

5 steps

AI, control and sovereignty

See how the enterprise keeps control of AI: a customer-controlled enforcement point, interoperable and reversible model choice, deterministic accountability, a governed learning system and the control plane above every harness.

  1. The Control Plane Must Be Yours
  2. AI Must Be Interoperable and Reversible
  3. AI Is Probabilistic. Accountability Is Not.
  4. Reality-Augmented Intelligence
  5. The Layer Above Every Harness
Begin this path
5 steps

Cyber, risk and resilience

Explore how Data Mediation creates one governable control surface across AI agents, IT and OT, enables fine-grained risk management, combines deterministic controls with probabilistic intelligence, turns new cyber knowledge into immediate protection and puts ISO/IEC 27090 into operation in the data path.

  1. Turning hybrid into singular
  2. Data Mediation for risk management: turning the impossible into the possible
  3. I’m 100% sure that we’re 67% secure
  4. The Zero-Day Inversion
  5. ISO/IEC 27090: putting AI security guidance into operation
Begin this path
3 steps

Origins and foundations

Trace the work from fraud control to Data Mediation and explore the research foundations behind the current architecture.

  1. From fraud control to Data Mediation
  2. Agile Delivery in the AI Era
  3. Foundational work, written for the present
Begin this path

Conversation · Cyber, risk and resilience

Cyber Uncut · 25 minutes

TomorrowX’s disruptive approach to legacy system protection

Kostas Siourthas, Deloitte Senior Partner Rob Parker and Major General (Retd) Dr Marcus Thompson AM discuss data protection, heritage-system risk and why protecting data in motion creates a practical alternative to disruptive replacement.

Research agenda

Questions guiding the next stage of the work.

How can AI gain useful access to enterprise systems without inheriting their risk?

Where must the AI control plane sit if models and SaaS services are not to become the enterprise’s enforcement authority?

How can protocol intelligence reduce the cost of connecting old and new?

How should operational evidence follow a solution from requirement through repair and retirement?

How can detection intelligence become immediate prevention while application teams retain time to patch, modernise or retire safely?

Take the work further

Move from the research into the Composable Agentic Platform.

Explore the Programmable Data Agent, Editor, Console and the lifecycle for composing and operating Data Mediation solutions.

Explore the platform