According to the World Economic Forum (WEF)’s Global Risks Report 2025, “adverse outcomes from AI technologies” was ranked as the sixth most severe risk over the next decade by business leaders and experts. Therefore, organizations must focus on designing their AI infrastructure and systems with security in mind to avoid these adverse outcomes and make the most of AI without cybercriminal-centric side effects.
In this context, it is encouraging to see the UK Government signalling ambition to prioritise security within the AI ecosystem via the introduction of a new AI Cyber Security Code of Practice.
Designed to guide developers, operators, and stakeholders in securely managing AI systems, the code of practice is divided into 13 core principles that cover every stage of the AI lifecycle, from design to deployment, maintenance, and end-of-life.
Among these, the Government has highlighted the need to “design your AI system for security as well as functionality and performance” in Principle 2. In addition, other key focus areas include enabling human responsibility for AI systems (Principle 4); securing your supply chain (Principle 7); documenting data, models and prompts (Principle 8); maintaining regular security updates and patches (Principle 11); and ensuring proper data and model disposal (Principle 13).
This code of practice doesn’t just apply to software vendors offering AI services to customers. It applies to any UK-based organization developing its own AI systems for in-house use, as well as any company using third-party AI components.
That breadth of application is important. Ultimately, the AI Cyber Security Code of Practice is about futureproofing. By building AI security and safeguards now, organizations will be better placed to prevent major problems down the road while still reaping the benefits of AI implementation, from greater productivity to improved decision-making.
In the UK NCSC’s own words, these principles in the Code of Practice are designed to “not only help enhance the resilience of AI systems against malicious attacks but foster an environment in which UK AI innovation can thrive”.
Of course, building resilience will be an ongoing journey rather than a destination, with cybercriminals evolving and adapting their techniques and targets all the time. However, already we have seen a variety of ways in which AI systems can be exploited if they are not designed securely.
Prompt injection attacks, for example, can enable malicious actors to bypass built-in safety guardrails to abuse large language models (LLMs) for nefarious means, and the dark web is reportedly already awash with ‘jailbreak-as-a-service’ offerings. Equally, misconfigurations in AI system components may be leveraged by threat actors seeking to steal or ‘poison’ sensitive training data or models. Such security holes have been discovered in vector databases, open-source components and LLM-hosting platforms.
These two avenues represent just a drop in a much broader ocean of potential threats to AI systems. If not managed correctly, these technologies can expand attack surfaces exponentially, increasing organizations’ risk of succumbing to attacks as they are increasingly embedded into key processes and applications.
That is why the AI Cyber Security Code of Practice is so crucial. If organizations are looking to ramp up their adoption of AI systems, then it is vital that they have the guidance and tools available to improve security, increase cyber resilience, and combat threat actors growing efforts to exploit malicious opportunities.
How ISO 42001 can help businesses adhere to the code of practice
Unlike the EU’s more prescriptive EU General-Purpose AI (GPAI) Code of Practice, the UK’s principles-based model applies technology-neutral regulations to AI – a reflection of the UK Government’s view that while regulating AI is necessary, acting prematurely may do more harm than good.
As a result, the terms of the AI Cyber Security Code of Practice remain somewhat open to interpretation and are likely to be refined moving forward. However, for businesses that are looking to get ahead, the good news is that there are already best practice frameworks and standards that AI developers and users can align with.
Enter ISO 42001 – a standard published in 2023 that has been specifically designed to help drive the responsible use and management of AI. With the goal of aiding organizations in establishing, implementing, maintaining and continually improving an AI management system, it provides guidelines on secure AI use, helping developers implement mechanisms to detect unusual behaviour or unexpected outputs, reducing susceptibility to manipulation.
In essence, ISO 42001 is an effective methodology for helping organizations to adopt AI mindfully. It advocates due diligence and caution, ensuring that there is sufficient insight and oversight of any risks associated with AI system development. And perhaps more importantly, the clauses and controls within ISO 42001 can be mapped to all the principles within the AI Cyber Security Code of Practice.
As the Code of Practice matures and evolves, it could become the de facto baseline for AI security. But for now, organizations can look towards ISO 42001 as a framework for best practice, guiding firms in managing and mitigating AI-specific security and operational risks, and enhancing the overall security posture of AI.
The author
Sam Peters, is Chief Product Officer of ISMS.online
