Building Responsible AI: KORE's Journey Towards ISO/IEC 42001

Share
Building Responsible AI: KORE's Journey Towards ISO/IEC 42001
Photo by Vitaly Gariev / Unsplash

As artificial intelligence becomes increasingly integrated into enterprise technology, the challenge is no longer simply determining what AI can do. Organisations must also establish how intelligent systems should be governed, secured and held accountable. At KORE Cloud Solutions, this challenge is shaping both our internal AI development programme and the evolution of our Business Management System.

Artificial intelligence is moving rapidly from experimental applications into the operational environments on which businesses depend. Language models are increasingly capable of interpreting technical information, analysing complex problems and interacting with external systems, while advances in agent-based architectures are making it possible to coordinate activities that previously required extensive manual intervention. For organisations operating cloud infrastructure, these developments create opportunities to improve service delivery, accelerate technical investigations and make more effective use of the considerable information generated by modern computing environments.

However, the introduction of AI into business-critical operations also creates responsibilities that cannot be addressed through technical capability alone. An intelligent system may correctly identify a network problem while simultaneously proposing an inappropriate corrective action. A model connected to operational databases may provide useful analysis while creating unacceptable risks if customer information is not properly isolated. Even a system that performs consistently during development may behave differently when confronted with incomplete information, unexpected failures or deliberately manipulated inputs.

These concerns are not reasons to avoid artificial intelligence, but they demonstrate why its adoption must be accompanied by sound engineering and effective organisational governance. As AI becomes more capable, the distinction between what a system can do and what it should be permitted to do becomes increasingly important. The surrounding architecture must enforce that distinction, while the organisation itself must establish responsibility for assessing risks, approving changes, monitoring performance and responding when something goes wrong.

At KORE Cloud Solutions, we believe that responsible AI development requires this combination of technical controls and management accountability. Our interest in ISO/IEC 42001 reflects that position and forms part of a broader programme to develop our existing management framework into an integrated Business Management System capable of supporting information security, business continuity, artificial intelligence governance and relevant regulatory requirements.

Why ISO/IEC 42001 Matters

ISO/IEC 42001:2023 is the international standard specifying requirements for an Artificial Intelligence Management System, commonly referred to as an AIMS. It provides a framework through which organisations can establish policies, responsibilities, objectives and processes for managing AI systems throughout their life cycle. Rather than prescribing particular algorithms or technologies, the standard concentrates on how organisations govern the development, provision and use of artificial intelligence, including the identification of risks, assessment of potential impacts and continual evaluation of management effectiveness.

This approach is particularly relevant because AI introduces characteristics that differ from those of conventional deterministic software. A traditional application generally executes operations according to explicitly defined program logic, whereas an AI model may produce different outputs depending on its training, configuration, context and the information presented to it. Although this flexibility is central to the usefulness of modern AI, it also creates challenges involving reliability, explainability, reproducibility and operational control.

An AI system may generate an answer that appears technically convincing without being factually correct. It may misinterpret information retrieved from an external source or fail to recognise that its available evidence is insufficient. Where models are connected to tools capable of performing real-world actions, such errors can have consequences beyond the generation of inaccurate text. They may affect infrastructure availability, information confidentiality or business processes.

ISO/IEC 42001 provides a management framework for addressing these concerns systematically. It encourages organisations to define the intended purpose of AI systems, identify relevant stakeholders, assess risks and impacts, establish appropriate controls and monitor whether those controls remain effective. It also requires organisations to consider the wider context in which AI is developed and used, including responsibilities, resources, competence and continual improvement.

For KORE, the attraction of this approach lies in its emphasis on repeatable management practices rather than one-off technical assessments. We want our AI projects to develop within an environment where expectations are understood from the beginning, where operational authority is explicitly controlled and where decisions about deployment are supported by evidence. Our objective is to work towards ISO/IEC 42001 certification as a demonstration of a functioning management system, rather than treating certification as the starting point for governance.

Building on Our Information Security Foundation

KORE has already achieved certification to ISO/IEC 27001:2022 for our established information security management system. This provides an important foundation for our wider governance ambitions because many of the disciplines required for responsible AI development are already familiar within an information security context. Risk assessment, asset management, access control, incident management, supplier evaluation, change management and internal auditing are all established components of a functioning information security management system.

The introduction of AI does not remove the need for these controls. In many respects, it makes them more important. An AI application connected to a virtualisation platform must still operate within defined access permissions, while a system retrieving customer information must respect the same confidentiality requirements as any other business application. Operational changes proposed or initiated through intelligent automation must remain subject to appropriate authorisation, logging and review.

Nevertheless, information security and AI governance are not identical. A system may be appropriately protected against unauthorised access while producing unreliable recommendations or creating unintended consequences for its users. A model may operate within a secure hosting environment while relying on unsuitable training data or exhibiting behaviour that has not been adequately evaluated. These issues require additional management considerations that extend beyond the traditional objectives of confidentiality, integrity and availability.

Rather than creating an entirely independent AI management structure, we intend to build on our existing framework and incorporate the additional requirements into a broader Business Management System. This is part of a wider evolution in how KORE manages its organisational responsibilities, particularly as our infrastructure footprint expands and our internal development programme introduces new technologies and services.

Our plans also include implementing a Business Continuity Management System aligned with ISO 22301, alongside the continued development of our information security arrangements across our infrastructure locations. By approaching these initiatives together, we aim to establish a consistent framework for governance while retaining the specific controls and evidence required by each standard.

Developing an Integrated Business Management System

The development of our Business Management System represents an opportunity to reconsider how management information is organised, maintained and used across the organisation. Traditional management systems can become fragmented when different standards are implemented independently, with separate policies, risk registers, audit schedules and management reviews covering many of the same underlying activities. Although such arrangements can satisfy individual requirements, they can also introduce unnecessary duplication and make it difficult to understand how risks and controls relate to one another.

Our intention is to develop a more integrated structure in which common management processes support several areas of governance. Information security, business continuity and AI management will retain their individual requirements, but activities such as document control, risk management, corrective action, internal auditing and management review can operate through a coordinated framework. This should make the system easier to maintain while providing a clearer understanding of how individual controls contribute to wider organisational objectives.

We are also reviewing the technology used to maintain our management documentation. Our existing ISO 27001 system has been supported through SharePoint, but we are exploring whether a more structured approach based on Markdown documentation, version-controlled repositories and linked management records could provide greater flexibility and traceability. This is particularly relevant to an organisation whose operational activities already involve software development, configuration management and technical documentation maintained through controlled engineering workflows.

The objective is not simply to move documents from one platform to another. We want to establish meaningful relationships between policies, procedures, operational assets, identified risks and the evidence demonstrating that controls are functioning. A policy governing access to an AI system, for example, should be connected to the technical mechanisms enforcing that access, the people responsible for maintaining those mechanisms and the records supporting their effectiveness. Changes to the underlying technology should be capable of triggering appropriate reviews of the associated documentation and risks.

Such an approach would also support our work on business continuity and regulatory preparedness, including requirements arising from the Digital Operational Resilience Act. While DORA, ISO 22301 and ISO/IEC 42001 address different concerns, they share an underlying need for clear accountability, accurate information, effective risk management and demonstrable operational control. A well-designed BMS can provide the common structure through which these responsibilities are coordinated without obscuring the distinctions between them.

Connecting Governance with Our Internal AI Projects

Our interest in AI governance is closely connected to the practical research and development taking place within KORE. Through our KLabs programme, we are exploring how intelligent systems can support infrastructure operations, technical analysis, security monitoring, software development and specialist business applications. These projects are at different stages of maturity and remain part of an evolving development programme, but they provide valuable opportunities to examine how AI can be integrated into real operational environments.

The wider KLabs ecosystem includes projects such as Oracle, Atlas, Sentinel, Forge and Nexus, each contributing to different aspects of our intended architecture. Oracle is being developed around AI orchestration and the coordination of intelligent capabilities, while Atlas explores the organisation and interpretation of infrastructure information. Sentinel contributes to our work on security-oriented analysis and operational visibility, with Forge and Nexus supporting development workflows and interaction with the wider platform.

Our longer-term intention is to establish an architecture in which specialised components can cooperate through controlled interfaces while maintaining clear boundaries between analytical reasoning and operational authority. Such an environment could eventually allow an engineer to investigate an infrastructure incident by drawing on monitoring information, network configuration records, virtualisation data and internal technical documentation without manually gathering every source of evidence.

The engineering challenge is ensuring that these capabilities operate safely. A model may be able to interpret routing information or propose a configuration change, but it should not automatically possess the authority to execute that change. Similarly, access to monitoring information must not provide an indirect route to confidential customer data. These restrictions must be enforced through the surrounding application and integration services rather than relying exclusively on instructions given to the model.

We are also exploring the use of locally hosted language models through GPU-based infrastructure, evaluating their performance, resource requirements and suitability for different tasks. Local inference may provide advantages where organisations require greater control over data processing, service dependencies and integration with internal systems, although hosting a model locally does not automatically make it secure or compliant. Appropriate evaluation, access controls, monitoring and lifecycle management remain necessary regardless of where inference takes place.

Another internal project, Solvera, explores AI-assisted financial market analysis and controlled paper-trading workflows. Although its intended application differs from infrastructure operations, it introduces related questions about the reliability of model-generated analysis, the boundaries of automation and the importance of human oversight. These considerations reinforce the value of establishing common governance principles that can be applied across different classes of AI system.

Translating Responsible AI into Engineering Practice

One of the most important principles guiding our intended architecture is that intelligence and authority should remain separate. Language models are useful because they can interpret information, identify relationships and propose actions, but they should not independently determine the security permissions under which those actions are executed. Authentication, authorisation, tenant isolation and operational policy enforcement must remain the responsibility of trusted software components.

This is particularly important in a multi-tenant cloud environment. KORE's infrastructure may support multiple customers whose systems and information must remain appropriately isolated. An AI component connected to operational systems must not be capable of retrieving one customer's information while responding to another customer's request. Such boundaries cannot be enforced reliably through conversational instructions alone; they require application-level access controls, appropriately scoped data retrieval and independent verification.

We are therefore considering an approach in which AI capabilities are introduced progressively. Read-only investigation and decision support provide opportunities to evaluate usefulness and reliability without granting models the ability to modify production infrastructure. More advanced automation can be considered where specific operational benefits have been demonstrated and where suitable approval mechanisms, execution controls and rollback procedures exist.

Standardised integration technologies, including the Model Context Protocol, may contribute to this architecture by providing consistent interfaces between AI applications and external systems. However, integration standards do not replace security engineering. Every exposed capability must be assessed according to its intended purpose, the information it can access and the consequences of its execution.

These design decisions are directly relevant to our ISO/IEC 42001 ambitions because effective AI governance must be reflected in the systems themselves. Written policies provide direction, but the technical architecture must demonstrate that those policies are enforceable in practice.

The Road Towards Certification

Our journey towards ISO/IEC 42001 certification will require a structured programme of implementation, evaluation and continual improvement. The first priority is to define the intended scope of our Artificial Intelligence Management System, identifying the organisational activities, AI systems, development processes and supporting services that will fall within its boundaries. This will need to reflect the systems we actually develop and operate, rather than simply the capabilities we hope to introduce in the future.

We will then need to assess our existing management arrangements against the requirements of ISO/IEC 42001, identifying which processes can be adapted from our established information security framework and where additional AI-specific controls are necessary. Particular attention will be required around AI system inventories, risk and impact assessment, lifecycle management, performance evaluation, human oversight and the management of changes to models and integrations.

As these processes are developed, our internal projects will provide practical opportunities to test their effectiveness. Applying governance requirements to real engineering work should help us identify where procedures are useful, where they introduce unnecessary complexity and where additional safeguards are required. We also intend to consider how the BMS can maintain the necessary evidence through existing development and operational workflows rather than depending entirely on manual administrative activity.

Before seeking certification, the management system will need to operate sufficiently to demonstrate its effectiveness. Internal audits, management reviews, corrective actions and documented performance evaluation will be essential parts of that process. An appropriately accredited certification body would then assess the system against the applicable requirements and agreed certification scope.

We have not established a public certification date. Our priority is to build a credible and sustainable management framework, with certification representing independent assessment of the system once it is ready.

Looking Ahead

Artificial intelligence is likely to become an increasingly important component of enterprise technology, but its long-term value will depend on the quality of the systems and organisational practices surrounding it. As models become more capable and are connected to a wider range of applications, questions of accountability, security, reliability and operational control will become more significant rather than less.

For KORE, the development of our Business Management System and our ambition to achieve ISO/IEC 42001 certification are part of preparing for that future. We want to create an environment in which intelligent systems can be developed and evaluated responsibly, where technical controls support clearly defined management objectives and where innovation can proceed without compromising the security or dependability of the services on which customers rely.

Our existing ISO 27001 certification provides an established foundation, while our work towards ISO 22301 and our consideration of DORA-related requirements will help extend the framework into wider operational resilience. The KLabs development programme provides an opportunity to apply these principles to practical AI engineering, ensuring that governance evolves alongside the technologies we are exploring.

There is considerable work ahead, and the development of both our management system and our AI platforms will be an ongoing process. Nevertheless, we believe that establishing the right foundations now will allow us to adopt new capabilities with greater confidence and maintain the level of accountability expected of a professional infrastructure provider.

Ultimately, responsible AI is not simply a matter of selecting an appropriate model or implementing a particular technical safeguard. It is about establishing an organisational approach in which intelligent technology is designed, operated and continually evaluated within a framework of clear responsibility and effective control. That is the direction we are pursuing at KORE, and it is the foundation on which we intend to build our future AI capabilities.

Read more

Beyond the LLM: Engineering Secure, Reliable AI Agents for Enterprise Infrastructure

Beyond the LLM: Engineering Secure, Reliable AI Agents for Enterprise Infrastructure

Building a capable artificial intelligence model is one challenge. Integrating that intelligence into business-critical infrastructure, while maintaining security, reliability and operational control, is an entirely different engineering problem. The rapid advancement of large language models has created considerable enthusiasm around the possibility of intelligent automation. Modern models can interpret

By KORE Pulse