This Government takes AI sovereignty and security extremely seriously.

We are investing £500m in UK AI firms via our SovAI Fund, including in home-grown model developers, like Ineffable. This goes beyond just funding, to include support such as visas and procurement.

Sovereignty is not just about access to AI models. It is about building leverage over key parts of the value chain, such as AI hardware. That is why the Government has announced a £1.1 billion AI hardware plan.

Security is also fundamental to the Government's approach. Our AI Security Institute is world leading, and one of a handful of organisations with access to Mythos before release.

This means the UK is not a bystander. Canada, France, Germany and elsewhere recognise our leadership, and want to work with us to build sovereign assets and learn from our approach to risks.

The statutory guidance on Best Value Standards and Intervention sets out the different forms of intervention and indicators of failure which inform a best value assessment. There are no powers within a Best Value Notice, as it is a non-statutory tool designed to set out the Department’s concerns and request further assurance from an authority.

The Tees Valley Combined Authority’s Best Value Notice sets out that the Authority’s progress will be reviewed after twelve months. The Notice remains in place until the Department confirms whether we deem it necessary to continue to seek assurance through such a Notice or not, or that further action is needed. The Department continues to require assurance of the Authority’s compliance with the Best Value Duty and Ministers are currently considering appropriate next steps based on the available evidence.

The Government recognises the importance of securing value for money for taxpayers and is committed to making the UK a world leader in AI, attracting private investment, supporting growth and creating high-quality jobs.

Commercial arrangements relating to specific projects are for the relevant parties.

The Government's world-leading AI Security Institute (AISI) has a dedicated red-team that tests the safeguards of frontier models. A summary of AISI's findings for Claude Mythos 5 and Fable 5 (a version of Mythos 5 with additional safeguards) are publicly available in the system card published by Anthropic.

Safeguards in AI models are improving, but they remain fallible, and companies have a clear responsibility to develop their products safely and to invest in the safety infrastructure this technology requires. The AI Security Institute continues to work with developers to improve model safety and address vulnerabilities, and that work will continue at pace as the technology evolves.

Decisions on AI products and services, including any contractual arrangements, are for individual departments to take. Information on individual contracts awarded above the relevant thresholds are published by individual departments.

The Department for Science, Innovation and Technology does not hold a central register of different department’s access to AI products and services, nor the contracts other departments have in place.

The Government utilises a range of suppliers based on operational requirements, value for money, and compliance with our security and legal obligations, with all suppliers subject to rigorous due diligence.

It is the responsibility of each contracting authority to ensure that contracts that they award are suitable for their requirements and legally compliant, and to monitor and manage the supplier's performance against their contractual obligations. All contracting authorities are simultaneously encouraged to follow the Government Security Group’s guidance on Tackling Security Risk in Government Supply Chains, which details best practices for procurement, commercial, and security practitioners when selecting and onboarding suppliers.

The Ministry of Defence (MOD) requires all suppliers, including Palantir, to meet UK Government security and supply chain assurance standards. Palantir remains subject to ongoing MOD commercial and security oversight, and the MOD is satisfied that appropriate assurances are in place.

The Government has signed Memoranda of Understanding with these companies to support industry cooperation. You can read the full details of these agreements on the GOV.UK website for each company: Anthropic, OpenAI, and Google DeepMind.

The Government is already working with these firms to deliver on UK objectives. For example, Anthropic are developing an AI assistant to transform GOV.UK services, and OpenAI are collaborating with the Ministry of Justice to support businesses with AI innovation.

Our AI Security Institute also collaborates closely with these companies to research the serious risks that advanced AI could pose. Through this collaboration, the Institute helps to identify vulnerabilities in developers' models; both OpenAI and Anthropic have addressed dozens of such issues before releasing their models.

The pilot of the GOV.UK AI assistant is currently being undertaken by GDS and will be concluding soon. We will then evaluate the concept to inform future development of the concept.

At this stage, the pilot is about learning and understanding the potential value of the technology for citizens in helping them get through complex life events. We will be considering accuracy and value as part of the final reporting on this pilot.

The Department does not hold any contracts with Anthropic.

The Department has continued to deepen its collaboration with Anthropic as part of its broader strategy to strengthen the UK’s sovereign AI capabilities. This partnership forms part of the Government’s commitment to ensuring the UK is an AI maker, not just a taker. Through the Sovereign AI Unit, DSIT is working with Anthropic and other frontier AI companies to shape the development of strategically important capabilities, leveraging the UK’s strengths in talent, data and investment to secure economic and technological advantage.