University of Surrey warns of UK cyber security risks from frontier AI export controls
The Growing Threat of Frontier AI Access
The rapid evolution of frontier AI systems has created a new dimension of risk for the United Kingdom. Recent actions by the United States government, which required licences before Anthropic could release its most advanced models to any foreign user, demonstrate that access to these models can be abruptly restricted. When a service is withdrawn, organisations that depend on it for network defence, data analysis, or critical operations may find themselves suddenly unable to operate. This volatility is not limited to government‑driven restrictions; fragile business models and heavy investment requirements make many frontier AI providers vulnerable to market forces, raising the possibility of service discontinuation for commercial reasons.
For the UK, where critical infrastructure spans energy grids, transport networks, and health services, the reliability of AI‑driven tools is becoming a matter of national security. The University of Surrey’s recent white paper, authored by Professor Alan Woodward of the Surrey Centre for Cyber Security and Dr Andrew Rogoyski of the Surrey Institute for People‑Centred AI, outlines five pragmatic recommendations that address these concerns without demanding the construction of a domestic frontier AI capability, which would be prohibitively expensive for a mid‑sized nation.
Why Sovereign AI Control Matters for UK National Security
National security strategies traditionally focus on physical defence and diplomatic relations. The emergence of AI as a dual‑use technology – capable of both attacking and defending digital networks – requires a shift in thinking. If a foreign government can halt access to a frontier model that a UK operator relies on, the ability to protect essential services could be compromised overnight. The paper points out that the United States has already shown it can enforce a “kill switch” through export controls, and that similar mechanisms could be applied by other states, including China. This reality makes it essential for the UK to retain control over the AI tools that safeguard its infrastructure.
Recent US Export Control Actions
In June 2026, the US Department of Commerce mandated licences for the release of Anthropic’s two most advanced models to any foreign person worldwide. Because the company could not reliably separate domestic from international users, it temporarily disabled the models for all users until the licences were processed on 1 July. This was the first instance where export controls directly impacted a live AI service, moving beyond the traditional scope of regulating hardware such as high‑end processors. The episode highlighted how quickly a policy decision can translate into a loss of access for allied nations, underscoring the need for resilient, sovereign AI strategies.
Implications for UK Critical Infrastructure
Frontier AI is increasingly being adopted by attackers to automate reconnaissance, exploit unpatched vulnerabilities, and launch large‑scale phishing campaigns. At the same time, defenders are using the same models to analyse traffic, detect anomalies, and automate response. If access to these models is cut off, both sides lose a powerful capability, potentially creating a security gap that adversaries can exploit. The white paper argues that the UK must therefore treat AI access as a strategic asset, recording the risk of loss in national risk registers and procurement terms, and negotiating access guarantees between governments rather than relying solely on commercial contracts.
Key Recommendations for UK Policymakers
Deploy Frontier AI Locally to Counter the “Foreign Kill Switch”
Rather than attempting the impossible goal of producing a domestically built frontier AI – a task that would require billions of pounds in hardware, specialised talent, and years of research – the paper recommends focusing on deploying existing frontier models within the UK’s own data centres. By hosting the models locally, the UK can enforce its own security boundaries, apply British data‑protection laws, and maintain continuity of service even if a foreign government imposes export restrictions. This approach aligns with the principle of “sovereign by necessity” and reduces reliance on external providers.
Record AI Access Risks in National Risk Registers
Embedding the probability of AI service loss into the UK’s national risk registers and procurement frameworks forces relevant agencies to consider this risk in their planning. When a supplier’s ability to provide AI tools is uncertain, the procurement process should flag the exposure, trigger contingency plans, and encourage the development of alternative solutions. This proactive stance helps ensure that critical infrastructure operators are not caught off‑guard when a model becomes unavailable.
Negotiate Government‑to‑Government Access Guarantees
Commercial contracts often lack the flexibility needed to guarantee continued access. The paper suggests that the UK government should negotiate bilateral or multilateral agreements with AI‑producing nations, securing written commitments that allies will receive advance notice and exemptions before any access restrictions are imposed. Such diplomatic channels mirror existing supply‑chain assurances in energy and defence, providing a more stable framework for AI services.
Invest Realistically in AI Capability and Talent Pipeline
While building a full‑scale frontier AI lab is unrealistic, the UK can invest in a realistic capability that leverages its existing research strengths. The paper emphasises the importance of a robust talent pipeline nurtured by universities such as the University of Surrey, which produce graduates skilled in AI engineering, cybersecurity, and data science. Targeted funding for scholarships, industry‑academia partnerships, and research centres can sustain this pipeline without the need for massive, one‑off infrastructure expenditures.
Maintain Focus on Fundamental Cyber Security Practices
Even with advanced AI tools, the baseline security controls – patch management, strong authentication, network segmentation, and regular vulnerability scanning – remain the most effective defence against cyber‑attacks. The authors note that AI can accelerate the discovery of existing weaknesses, but it does not replace the need for solid security hygiene. For a country with limited resources, spending on these fundamentals yields a higher return on protection than funding only cutting‑edge AI research.
Diplomacy and Policy Actions Needed
Beyond technical measures, the white paper calls for a coordinated diplomatic effort. The UK should press Washington for transparent criteria that determine when access to frontier AI will be restricted, seek advance warning of any changes, and negotiate exemptions for allied nations. By treating AI access as a matter of international policy, the UK can avoid the costly duplication of capabilities that would otherwise be required to compensate for sudden service loss.
Leveraging UK Talent and Research Strength
Britain’s greatest advantage in the AI arena is its world‑class academic ecosystem. Institutions such as the University of Surrey have long‑standing expertise in computer science and cybersecurity, and they have produced many of the researchers who now work in leading AI labs abroad. By strengthening collaborations between these universities, national research institutes, and industry partners, the UK can create a vibrant ecosystem that extracts maximum value from existing frontier models while developing home‑grown expertise. This approach not only supports national security but also sustains economic growth and innovation.
Conclusion: Taking the Next Step
The convergence of export‑control actions, the rise of AI‑enabled cyber threats, and the fragility of commercial AI business models has created a pivotal moment for UK cyber security policy. The white paper from the University of Surrey provides a clear, actionable roadmap: secure local deployment of AI, formalise risk recording, pursue diplomatic access guarantees, invest wisely in talent, and keep security fundamentals at the forefront. Implementing these steps will enable the UK to maintain resilient, sovereign control over the AI tools that protect its critical infrastructure, without the unrealistic expectation of building an indigenous frontier AI ecosystem.
Schedule a free consultation with Surrey experts
Subscribe to our newsletter for AI security updates
Read related articles on AI and cyber security
