AI Governance for Defence: ISO 42001 in Sovereign Security Environments

The Ministry of Defence's approach to AI has evolved rapidly. The Defence AI Strategy of 2022 established AI as a strategic priority. The subsequent development of the Responsible AI Principles, the establishment of the Defence AI Centre, and the integration of AI governance requirements into defence procurement frameworks have created an environment in which AI governance is not a desirable attribute for defence suppliers - it is a threshold requirement.

ISO 42001, published in December 2023, provides the international standard for AI management systems. In most commercial and public sector contexts, it provides a complete and adequate governance framework. In defence contexts, it provides the foundation - but the defence-specific requirements extend beyond what the standard addresses.

The classified environment creates additional requirements. AI systems that process classified information, or that produce outputs that would themselves be classified, require governance mechanisms that operate within the classification framework. The evidence base for ISO 42001 compliance - the documentation of risk assessments, impact analyses, and control implementations - may itself be classified, creating challenges for the independent audit process that certification requires. Certification bodies operating in classified environments require specific authorisation. The number of accredited certification bodies with this capability is small.

Export controls present a further consideration. The development and export of AI systems with military applications may be subject to export control regulations - the UK's Export Control Order, US ITAR and EAR, and the emerging EU AI Act dual-use provisions. AI governance documentation that describes the capabilities and characteristics of AI systems with military applications requires careful management within these frameworks. The supply chain transparency that ISO 42001 requires - and that is generally desirable - may conflict with export control restrictions on the disclosure of technical information.

Sovereign capability requirements add a further layer. The UK's defence AI strategy explicitly identifies sovereign capability - the ability to develop, operate, and maintain AI systems without dependence on foreign capability - as a strategic objective. Governance frameworks for defence AI need to address not just the management of existing AI systems, but the development and maintenance of the institutional capability required to govern those systems over time, within the UK's sovereign capacity.

Building a defence-appropriate AI governance architecture requires integrating ISO 42001's management system requirements with the specific requirements of classified environments, export control frameworks, and sovereign capability objectives. It is a more complex undertaking than ISO 42001 certification in commercial contexts. It is also more consequential - both for the individual organisation and for the UK's defence AI capability.