UK government launches AI Biosecurity Lab with £100m and up to £200m in OpenAI and Anthropic credits
The UK government has announced the AI Biosecurity Lab (AIBL), an independent non-profit organisation that will use frontier AI models to detect emerging biological threats and accelerate the development of vaccines, treatments and other medical countermeasures. The Cabinet Office describes it as the world's first dedicated AI biosecurity organisation. The government is putting £100m of public money into the lab, while OpenAI and Anthropic have committed up to £200m in compute and model credits, trusted access to their most capable relevant models and, in Anthropic's case, dedicated technical support.
The lab was announced on 24 October 2026 in a Cabinet Office release, and HM Treasury placed it the same morning inside a wider AI package presented by Chancellor John Healey ahead of his Budget. That package also includes an additional £120m for the AI Security Institute, a £60m National Cyber Shield built around agentic AI for defence, and a £70m National Exposure and Real-Time Vulnerability (NERV) Centre that will deploy AI tools to find vulnerabilities across critical infrastructure. The security measures sit next to investments in sovereign AI companies, chip design skills, regional AI partnerships and a new AI Economics Institute.
The announcement matters less for its headline sums, which are modest by the standards of frontier AI spending, than for the institutional model it sets up. Britain is creating a public-interest organisation whose main technical inputs are frontier models supplied under controlled access by the companies that build them, and it is pointing that organisation at a domain where the same models are treated by their developers as a potential source of risk.
··········
HOW THE £165M NATIONAL AI BIOSECURITY INITIATIVE IS STRUCTURED.
The lab is the largest component of a £165m National AI Biosecurity Initiative. According to HM Treasury, £100m goes to the AIBL itself, while the remaining £65m is earmarked to strengthen biosurveillance and to develop medical countermeasures for priority pathogens. The government says it also intends the lab to attract further public and philanthropic funding over time, which suggests the £100m is meant as founding capital rather than a complete multi-year budget. No spending profile by year has been published.
The private commitments are in kind rather than in cash. OpenAI has pledged £100m in compute and model credits together with trusted access to its most capable relevant models through its Rosalind Biodefense programme. Anthropic has pledged up to £100m in model credits, trusted model access and dedicated technical support. Two further partners contribute expertise rather than money: Google DeepMind will offer research collaborations with its scientists and biosecurity specialists and, according to the Cabinet Office, potentially trusted access to Gemini and specialised science models, while Basecamp Research, a London-based AI life sciences company, will provide access to its AI-biology models and technical staff.
........
Component | Amount | Form |
AI Biosecurity Lab (government) | £100m | Public funding |
Biosurveillance and countermeasures | £65m | Public funding |
OpenAI commitment | £100m | Compute and model credits |
Anthropic commitment | Up to £100m | Model credits and support |
Google DeepMind | Not quantified | Research collaboration |
Basecamp Research | Not quantified | AI-biology models and expertise |
........
The headline figure of "up to £300m" for the lab therefore combines £100m of government money with up to £200m of vendor credits whose real value depends on list prices, usage and the terms of access. Credits are useful for an organisation that will spend much of its budget on inference, but they also tie a share of its capacity to specific providers. The government has not said how the credits will be allocated across projects, whether they expire, or how the lab would replace them if a partner's commitment changes.
··········
AN INDEPENDENT NON-PROFIT MODELLED ON THE AI SECURITY INSTITUTE.
The AIBL is set up as an independent non-profit organisation rather than as a unit inside a ministry. The Cabinet Office says it is modelled in part on the AI Security Institute, the government body that evaluates advanced models before and after release, but with the autonomy to work "at the pace of a frontier AI-bio company". A scientific advisory committee drawing on academia, industry and the third sector will advise it, with membership to be announced later. Leadership, headcount, location and an operational start date have not been disclosed.
The lab will sit inside a National AI Biosecurity Initiative that also brings together the UK Health Security Agency (UKHSA), the Animal and Plant Health Agency (APHA), the Defence Science and Technology Laboratory (Dstl) and the life sciences sector. The Medicines and Healthcare products Regulatory Agency (MHRA) will work with the AIBL on health emergency preparedness and on the regulation of AI-designed medicines. That last point is notable: it places the medicines regulator inside the programme from the outset, rather than leaving the question of how AI-designed countermeasures are assessed until a candidate is ready for trials.
The government frames the lab as an extension of the UK Biological Security Strategy. Its stated mission is to build practical technical capabilities to understand, prevent, detect and respond to biological threats, explicitly including AI-enabled ones. The two initial focus areas are early detection of emerging biological threats and faster development of vaccines, treatments and other countermeasures.
··········
WHAT THE OPENAI AND ANTHROPIC MODEL ACCESS LIKELY INVOLVES.
The phrase "trusted access" carries most of the technical weight in the announcement. Frontier developers increasingly restrict biology-relevant capabilities in their public products and release them more fully only to vetted users under additional controls. OpenAI's Rosalind Biodefense programme, which the Cabinet Office names as the channel for its UK commitment, was introduced by OpenAI in May 2026 to help trusted developers build biodefense and pandemic preparedness tools. At that time OpenAI also expanded trusted access to GPT-Rosalind, which it describes as a frontier reasoning model built for life sciences research, for selected US government and allied partners with public health and biodefense missions.
OpenAI's own description of that programme lists the kinds of work it expects such access to support: epidemiological modelling, early detection, DNA synthesis screening, preparedness planning, non-pharmaceutical interventions and medical countermeasure development. It also describes the safeguards wrapped around the access, including bio-specific capability assessments, safer model behaviour for dual-use biological requests, monitoring and enforcement, expert red teaming and security controls for higher-risk capabilities. Government access, according to OpenAI, is tied to safety, security and accountability controls. The UK announcement does not specify which OpenAI models the AIBL will use, so it is an inference, not a disclosed fact, that GPT-Rosalind or its successors will be among them.
Anthropic's commitment is described only in general terms: up to £100m in model credits, trusted model access and dedicated technical support. The government has not said which Claude models are involved or under what usage policies. The same applies to Google DeepMind, whose participation is framed as research collaboration with possible trusted access to Gemini and specialised science models. For a lab whose purpose is to work near the boundary of dual-use biology, the access terms, logging requirements and audit rights will determine much of what it can actually do, and none of them has been published.
··········
THE £120M AI SECURITY INSTITUTE TOP-UP AND THE CYBER MEASURES.
The biosecurity initiative is the most novel part of the announcement, but HM Treasury grouped it with three other security measures that together account for most of the package's protective spending.
........
Measure | Amount | Stated purpose |
National AI Biosecurity Initiative | £165m | AIBL plus biosurveillance and countermeasures |
AI Security Institute (additional) | £120m | Testing, research and technical standards |
National Cyber Shield | £60m | Agentic AI for cyber defence |
NERV Centre | £70m | AI tools to find vulnerabilities in critical infrastructure |
........
The £120m top-up for the AI Security Institute is earmarked for testing, research and technical standards. In a speech on 12 October, AI Minister Kanishka Narayan said the institute employs more than a hundred technical experts and that its scientists evaluate leading models before release, citing Gemini 4 Argon as an example. He also said Britain is exploring hardware-enabled verification, which he described as early-stage research.
The £60m National Cyber Shield is intended to use agentic AI to improve cyber defences. It includes a £25m challenge fund for UK AI and cyber security companies and scoping work for a regulatory sandbox in which agentic security tools could be tested across critical national infrastructure, the public sector and industry. The sandbox is at the scoping stage only, and no timetable or participating operators have been named.
The £70m NERV Centre is designed to deploy AI tools that find cyber vulnerabilities across critical infrastructure and the wider economy. It builds on two government pilots that the Cabinet Office says used frontier AI to identify and fix more than a thousand potential vulnerabilities across government systems. In his 12 October speech, Narayan added that a central vulnerability-monitoring service had cut the time to fix weaknesses from 50 days to eight. The government has not published the methodology behind either figure, so they should be read as official claims rather than independently verified results.
··········
THE REST OF THE PACKAGE: SOVEREIGN AI, CHIPS AND ECONOMIC RESEARCH.
The security spending is presented alongside measures aimed at building domestic capability and spreading adoption. The Sovereign AI Fund receives an additional £60m to invest directly in UK AI companies, with a new focus on physical and industrial AI covering robotics, autonomous systems and advanced manufacturing. Another £60m goes to the Sovereign AI R&D Procurement Scheme, which buys solutions in health and care, AI security, assurance and safe adoption. A further £18m funds a UK Advanced Chip Design Academy intended to build chip-design skills and connect research with semiconductor manufacturing, and the Treasury also refers to new investment in AI chip design capability in Edinburgh.
On the economic side, the AI Economics Institute receives £23m over three years, and the government commits £10m a year to work on how AI will reshape jobs and industry. A £15m AI Challenge Prize will reward AI applications that solve common problems, with health as the first target, and £30m will launch local AI partnerships with Mayoral Strategic Authorities. The National Cyber Security Centre will also open an AI Cyber Security Centre of Excellence in Manchester to train government cyber and AI specialists, with no amount attached.
The multi-year line items published by HM Treasury add up to £621m, excluding the recurring £10m a year and the in-kind company credits, which is consistent with the "£600m+" description used by techUK in the government's release. The Treasury also cited two public-service examples to illustrate adoption: AI tools that help clinicians analyse more than 210,000 NHS chest X-rays a month, and an HMRC digital assistant that handled around 6.3 million customer interactions in 2025-26.
··········
WHY BIOSECURITY IS THE AREA WHERE FRONTIER LABS ARE MOST CAUTIOUS.
Biology is one of the domains in which frontier developers apply their strictest internal controls. OpenAI states that ChatGPT agent, released in July 2025, was the first model it treated as High capability in biology under its Preparedness Framework, and both OpenAI and Anthropic now route the most sensitive biology use cases through vetted-access programmes rather than general availability. The AIBL is effectively a state-sponsored channel through which those restricted capabilities can be applied to defensive work, with public oversight and with the lab rather than individual researchers as the accountable party.
That arrangement has clear technical advantages. Early detection of novel pathogens depends on combining genomic surveillance data, epidemiological signals and expert interpretation at speed, and countermeasure development increasingly relies on protein and sequence models that can propose candidate antigens or binders long before laboratory validation. A lab with privileged access to several frontier models, specialised biology models from Basecamp Research and data relationships with UKHSA and APHA could, in principle, shorten both loops. CEPI, the vaccine coalition whose deputy chief executive is quoted in the Treasury release, has organised its strategy around a target of developing vaccines within 100 days of a new threat being identified, and the lab is presented as complementary to that goal.
The same arrangement also concentrates sensitive capability. Trusted access to models that are deliberately restricted for the public creates obvious requirements for personnel vetting, data handling, logging and incident reporting, and the government has not yet described how the AIBL will meet them. The MHRA's role covers regulation of AI-designed medicines, but there is no published framework yet for how the lab's own outputs, such as sequence designs or threat assessments, will be reviewed before they are shared.
··········
WHAT IS ESTABLISHED AND WHAT REMAINS OPEN.
The established facts are specific. The UK has created an independent non-profit AI Biosecurity Lab with £100m in public funding, inside a £165m National AI Biosecurity Initiative that also funds biosurveillance and countermeasures. OpenAI has committed £100m in compute and model credits plus trusted access through Rosalind Biodefense, Anthropic up to £100m in credits plus trusted access and technical support, and Google DeepMind and Basecamp Research have agreed research partnerships. The lab is modelled in part on the AI Security Institute, will be advised by a scientific committee, and will work with UKHSA, APHA, Dstl and the MHRA. Alongside it, the AI Security Institute receives an extra £120m, a £60m National Cyber Shield introduces a £25m challenge fund and a planned sandbox for agentic defence tools, and a £70m NERV Centre will apply AI to vulnerability discovery in critical infrastructure.
The open questions are mostly operational. The government has not named the lab's leadership, staffing level, location or start date, nor has it published the terms of model access, the expiry and allocation of the credits, or the governance rules for handling dual-use outputs. The figures on vulnerabilities found by AI in government pilots come without methodology. Chancellor John Healey is expected to set out further details in his Budget, and those details, together with the composition of the scientific advisory committee, will show whether the AIBL becomes a working technical organisation with real model access or remains primarily a funding commitment.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
