On October 24, 2026, ahead of the Budget, the UK Treasury announced a large-scale AI investment plan that includes measures against biological threats and cyber defense. Four AI security programs will receive a combined £415 million, and separate funding will go toward developing the domestic AI industry and building talent. Building on the AI risk assessment and cyber defense pilots announced on October 12, the latest announcement sets out budgets and delivery bodies for each program. The government plans to invest not only in evaluating the dangers of AI, but also in defending against real cyberattacks and developing vaccines, therapeutics and other tools to prepare for biological threats.
£415 Million Allocated to Four AI Security Programs
The four programs the Treasury listed under "AI Security" in its October 24 announcement total £415 million. This figure is the sum of four independent programs' budgets; it does not double-count allocations to individual measures within each program.
| Program | Announced investment | Main purpose and budget breakdown |
|---|---|---|
| National AI Biosecurity Initiative | £165 million | Countering biological threats. £100 million to establish and run a laboratory; £65 million for surveillance systems and medical countermeasures against priority pathogens |
| Additional investment in the AI Security Institute (AISI) | £120 million | Evaluating the capabilities and risks of AI models and developing technical standards |
| National Cyber Shield | £60 million | AI-agent-based cyber defense and deployment support. Includes a £25 million competitive fund |
| NERV Centre | £70 million | Using AI tools to find vulnerabilities across infrastructure and the wider economy |
| Total | £415 million |
The National AI Biosecurity Initiative's £165 million includes £100 million for the laboratory and £65 million for surveillance and medical countermeasures. Likewise, the National Cyber Shield's £25 million competitive fund is part of that program's £60 million. The compute resources and AI model access offered by private companies differ in nature from government cash spending, so they are not included in this total.
The £415 million is the combined total for the four AI security programs, not the total for the entire AI investment plan. It is also not an annual budget, and the spending periods of the programs are not necessarily the same. Nor does it represent actual spending or defense capabilities already in place.
A notable feature of this allocation is that it divides AI security challenges into several areas and gives each its own budget. Preventing, detecting and responding to biological threats, evaluating the safety of AI models themselves, finding vulnerabilities in infrastructure and systems, and supporting the deployment of defensive technologies are each positioned as separate programs. The plan covers everything from understanding AI risks to building the capacity to respond to actual threats.
AI Biosecurity Lab Established, With Compute From OpenAI and Anthropic
At the center of the National AI Biosecurity Initiative is the newly established AI Biosecurity Lab (AIBL). According to the UK government's announcement of the lab, it will initially focus on detecting emerging biological threats early and accelerating the development of vaccines, therapeutics and other countermeasures. Its scope is not limited to threats created using AI but includes broader public health risks. The government explains that while AI advances scientific research, it may also lower the barrier to obtaining the knowledge and skills needed for harmful acts.
In addition to investing £100 million in the lab, the UK government will receive compute resources and model access for research from private AI companies. OpenAI will provide compute and model access worth £100 million and will grant access to relevant frontier models through Rosalind Biodefense. Anthropic also plans to provide model access worth up to £100 million, along with dedicated technical support. The government says it will grant access to both companies' models on the basis of trusted relationships with research institutions, but it has not published specific terms of use.
However, the support from both companies is not a cash donation. Model access refers to the right to use a company's AI models and compute resources within a defined scope, and the stated amount represents the value of the resources made available. Anthropic's figure of £100 million also carries the qualifier "up to." Simply adding the government's lab budget to the companies' access credits and treating the result as cash investment would misrepresent the nature of the support.
Other companies have also pledged cooperation with the lab. Google DeepMind aims to provide access to Gemini and science-focused models in addition to research collaboration, but this is not an announcement that access is already available. Basecamp Research plans to contribute its AI models, technical capabilities and expertise. The type and stage of each company's support differ, so the range of resources the lab can actually use will have to wait for further detail.
AIBL is planned as an independent nonprofit organization. The UK Health Security Agency (UKHSA), the Animal and Plant Health Agency (APHA) and the Defence Science and Technology Laboratory (Dstl) are involved in the National AI Biosecurity Initiative. It will also work with the Medicines and Healthcare products Regulatory Agency (MHRA) on health crisis preparedness and the regulation of AI-designed medicines. The setup connects research on AI-enabled biological threats with existing government bodies responsible for public health and drug regulation.
Much of the lab's operating structure remains undecided. The government says details of its scientific advisory board will be announced later, and the start date, location and head of the lab have not been disclosed. Year-by-year allocation of government funds, along with the terms of use and valuation methods for the models offered by companies, have also not been published. How the pledge to secure the necessary resources translates into actual research and the ability to respond to biological threats will depend on how the lab is run.
From AI Risk Assessment to Deploying Cyber Defense
On October 12, the UK government also announced measures on AI-enabled cyber defense and risk assessment of frontier AI models. That announcement said two pilot programs had helped identify and fix more than 1,000 vulnerabilities in government systems. It also set out approaches to evaluating frontier AI models and responding to AI-related incidents, and said stronger legislation would be considered. Note that the figure of more than 1,000 is the government's description of the earlier pilots, not a new count of vulnerabilities fixed as a result of the October 24 investment announcement.
The latest announcement makes the budgets and delivery bodies for these efforts clearer. Notably, it positions evaluating AI models themselves and defending systems using AI as separate programs.
The AI Security Institute (AISI) evaluates the capabilities and risks of frontier AI models and is also responsible for developing related technical standards. The NERV Centre, meanwhile, will use AI tools to find vulnerabilities in the systems that underpin critical infrastructure and economic activity. The National Cyber Shield strengthens cyber defense using AI agents, and envisions using technologies that carry out complex defensive tasks with minimal human oversight. In biosecurity, funding will go not only to early threat detection but also to medical countermeasures such as vaccines and therapeutics.
These programs differ not only in the risks they address but also in the work they carry out. Evaluating the dangerous capabilities of AI models is a different challenge from finding weaknesses in real infrastructure and systems. And even when vulnerabilities are found, additional work is needed to fix them and operate defenses in the field. The UK government is trying to allocate budgets to each of these stages: evaluation, discovery and defense.
For the National Cyber Shield, the government is also considering a "regulatory sandbox" to encourage the adoption of AI defense technologies by critical infrastructure, the public sector and industry. This is a framework that allows new technologies to be tested under certain conditions and supervision. At this point, however, it is only under consideration; the framework has not yet started. It is positioned as a mechanism for testing how defenses can be applied in real operating environments against vulnerabilities discovered by AI.
Just before the investment plan was announced, events suggesting the impact of AI-driven vulnerability discovery were also reported. According to an AFP report carried by Tech Xplore, government agencies and companies shut down websites on October 22 after a serious PHP vulnerability discovered by AI. However, no evidence was presented that this event directly determined the October 24 investment plan.
The announcement shows that the UK government emphasizes not only AI risk assessment but also everything from vulnerability discovery to deploying defenses. But specifics of operation remain unclear, such as what technical specifications will link the programs and how far the response to discovered vulnerabilities will be automated. It is not an announcement that automatic fixing of vulnerabilities by AI has been perfected, and the defensive benefits of the new investment have not yet been demonstrated.
Balancing a Domestic AI Industry With Use of Overseas AI Models
In addition to investing in security, the UK government announced measures to strengthen the competitiveness of its domestic AI industry. It will add £60 million to the Sovereign AI Fund, expanding support for AI technologies used in physical environments and industry, such as robotics, autonomous systems and advanced manufacturing. It also plans to allocate £60 million to government procurement for research and development and £18 million to a chip design academy. By combining funding for companies and research institutions with government technology procurement and training of semiconductor design talent, the aim is to strengthen the foundations of domestic AI development.
The government will also invest £15 million in the AI Challenge Prize and an additional £23 million over three years in the AI Economics Institute. A further £30 million is planned to strengthen links between regional AI organizations. The plan also includes talent development at the NCSC center in Manchester and investment in chip design in Edinburgh. These are not included in the total for the four security programs above, but they are measures to build domestic R&D capacity and skilled talent to support long-term use of AI.
The UK government says it aims to raise domestic AI development capacity and reduce dependence on overseas technology. At the same time, the AI Biosecurity Lab will make active use of models and compute from overseas companies such as OpenAI and Anthropic. This is not necessarily a contradiction. Judging from this plan, what the UK appears to prioritize is not developing all AI technology at home, but using the technology it needs while securing the domestic ability to assess and respond to threats.
However, simply gaining access to overseas companies' models does not establish a domestic AI security framework. What matters next is whether year-by-year budget allocations become clear and whether the AI Biosecurity Lab's operating plan and model-access terms are fleshed out. Moreover, turning threats discovered by AI into real medical responses and infrastructure defense will require mechanisms for putting each program's results to use in the field, including pilots and deployment conditions for the National Cyber Shield. In judging the results of this investment plan, the focus will be not only on the size of the budget but on how far each program can build practical defense and response capabilities.