AI and Defence: AI Practitioner’s Handbook

Earlier this year, the UK’s Defence AI Centre (DAIC) published the AI Practitioner’s Handbook (AIPH), a new guide designed to support the responsible and effective development of AI in the defence sector in accordance with the MOD’s policy framework, outlined in Joint Service Publication (JSP) 936. The AIPH is structured broadly around the AI system lifecycle, reinforcing the message that AI risks and obligations evolve over time and must be managed continuously. 

For those developing, deploying or supplying AI‑enabled systems to Defence, the AIPH provides an important indication of how MOD expects AI to be designed, evidenced and assured in practice.

This article explores the key themes of the AIPH for leaders, developers, users and suppliers who are responsible for AI assurance in the defence sector. 

Setting the strategic context for “ambitious, safe and responsible” AI (relates to JSP 936 Part 1, sections 1–2) 

The AIPH outlines the senior roles within an organisation which create the culture and systems to deliver “ambitious, safe and responsible” AI. This includes Responsible AI Senior Officers (RAISOs), who oversee responsible AI and are accountable for the implementation of AI ethics and compliance activities in their organisation. Notably, RAISOs are nominated in MOD organisations and industry partners are not expected to appoint a RAISO. However, many suppliers will have similar roles within their organisation, and effective engagement with MOD assurance processes will often require alignment with these functions.

Other roles considered to create an AI-enabled environment include Chief Technology Officer, Chief Data Officer, Chief Information Security office and training managers. These roles collectively support the organisational conditions in which AI can be developed and used responsibly.

More broadly, the AIPH groups roles involved in AI delivery according to their common remits, covering those involved in design, development, testing or implementation of AI systems. These groups are categorised as: 'Business', 'Technical' and 'User' groups. The AIPH highlights the importance of representation across these groups so that AI systems are tested from both operational and technical perspectives. For suppliers, this reinforces the importance of multidisciplinary engagement when developing and validating AI solutions for defence use.

Implementing and documenting AI in development stages (relates to JSP 936 Part 1, sections 5–6)

A core theme of the AIPH is the expectation that responsible AI considerations are embedded during development and clearly documented.

• Human centred design: the MOD requires those developing AI systems to consider harm and human factors for those affected by the models used. It is not enough to consider technical performance alone. Organisations should consider the range of positive and negative effects regarding useability, system effectiveness, end-user acceptability and a lower need for training. 

  For suppliers, this means demonstrating that human needs, limitations and risks have been actively considered throughout the AI lifecycle, from problem definition through to deployment and use.

• Model cards: the cards are living documents which help organisations understand how an AI model has developed by providing information on model-specific risk and use. The AIPH templates are short documents covering the purpose and performance of a model, together with model‑specific risks and limitations. These assist teams to track models with an auditable record and facilitate reuse. 

• System cards: Model cards alone won't provide sufficient visibility of AI risks or system context. Additional documents, such as data cards and system cards, are also needed to provide a holistic view of models. System cards provide a structured way for defence organisations to document and assess AI systems and deployed models. Their aim is to provide a 'single source of truth' for system-level risks and environment.

  For suppliers, this highlights the importance of system‑level assurance evidence, not just model‑level performance metrics. 

Assuring AI systems (relates to JSP 936 Part 1, sections 7–9) 

The AIPH offers a step-by-step guide to support industry in aligning with the MOD’s AI assurance approach. It emphasises that assurance is a cyclical process rather than a one‑off ‘tick‑box’ exercise, with risks and benefits requiring continuous review and update.

The AIPH sets out a clear process map of recommended activities, evidence requirements, and escalation routes across development teams, Senior Responsible Officers (SROs), RAISOs, and MOD Component Organisations. It also provides resources addressing the key stages of AI assurance, including project initiation, evidence gathering, deployment, testing, and ongoing monitoring.

By using the templates and checklists provided, suppliers can implement a robust, cyclical AI assurance process that meets required assurance standards.

Legal, ethics and governance (to relate to JSP 936, Part 1 Section 3 and 4) 

The AIPH flags dedicated content on legal, ethical and governance considerations, aligned with JSP 936, although this area is still in development at the time this article was written. 

Nevertheless, on the basis of JSP 936 standards legal and ethical compliance is intended to be embedded within AI assurance, rather than treated as a separate or purely advisory exercise. Suppliers should therefore anticipate increasing focus on how legal, ethical and governance risks are identified, managed and evidenced as the AIPH matures.

The AIPH, alongside JSP 936, the Defence AI Playbook and the Defence AI Strategy, offers practical guidance when implementing AI whilst maintaining strong governance. For those developing, deploying or supplying AI‑enabled systems to defence, the AIPH provides an important indication of how MOD expects AI to be designed, evidenced and assured in practice.

If you would like to discuss defence and how current or future regulations impact what you do with AI, please contact Lucy Owens and Louise Dean in our Defence Team, or Tom Whittaker, Brian Wong, Lucy Pegler, Martin Cook, Liz Griffiths or any other member in our Technology team.  For the latest on AI law and regulation, see our blog and newsletter.