Bird & Bird | Kate Deniston and Roger Bickerstaff

European Union | United Kingdom

Today, on 28 April 2026, UK Tech Minister Liz Kendall delivered a speech setting out her plans for AI in the UK.

Back in January, Liz Kendall commented on how the UK is ‘uniquely placed to set the standards for how AI is used and adopted’, later saying that ‘when it comes to AI standards and deployment, Britain can be a beacon to the world’. She then added that she would ‘say more about Britain’s role in the world on AI in the coming months.’

Today we learnt more about these plans.

What are the key announcements?

In her speech, Liz Kendall explained that there are two ‘key shifts’ in the government’s approach to AI in the UK:

  1. Going forward, the UK government will support more British AI companies, particularly in areas that the UK has strong expertise, such as AI hardware; and
  2. The UK government will work closely with other countries, particularly ‘other so-called middle power nations’, including on ‘setting the standards for how AI is deployed’.

On the first limb, Liz Kendall announced that she will launch a new ‘AI Hardware Plan’ at London Tech Week in June. Liz Kendall also contemplates the UK securing 5% of the global AI chips market.

However, in this article, we focus on the second limb.

What is meant by ‘setting the standards for how AI is deployed’?

Liz Kendall announced that:

‘in July, at the next meeting of the international network of AI Security Institutes, which the UK Chairs, we will publish best practice on the science of evaluating AI models.’

Established in November 2024, the international network of AI security institutes constitutes Australia, Canada, the EU, France, Japan, Kenya, the Republic of Korea, Singapore, the UK and the US.

The UK’s AI Security Institute, set up by the previous Sunak-led government, is widely regarded as being a leader in the field of AI safety testing. The UK holds the role of Network Coordinator of the network of AI Security Institutes and helps shape the efforts to rigorously measure and evaluate AI models.

It seems that the UK government’s plan for advancing the safe and responsible deployment of AI is to publish guidance for other AI security institutes to use when testing AI models. This would help other nations’ institutes to ensure that AI models being deployed in their jurisdiction undergo robust safety testing, presumably with the intention that this ultimately leads to a high standard of safety in AI tools being deployed globally.

Bird & Bird Analysis

The UK government is seeking to use the success of its AI Security Institute to help it lead on safe AI deployment globally. However, the government’s focus on safety testing AI models is not new.

At the inaugural AI Safety Summit back in November 2023, the UK’s Conservative government announced that it had coordinated a voluntary commitment between major AI developers and nations to safety-test models both pre- and post-deployment. When Labour came to power in July 2024, they announced through the King’s speech that they would introduce ‘appropriate legislation’ to place requirements on developers of the most powerful AI models. Statements made by Peter Kyle as shadow Tech Secretary before coming to power indicated that he was considering implementing a statutory code under which AI companies would be legally required to share testing data with government. AI companies would have to inform the government whether they were planning to develop AI systems over a certain level of capability and would need to conduct safety tests with independent oversight. The plan to introduce legally-binding rules on developers has since been shelved. However, almost two years later, we are still seeing the UK government focused on AI model safety testing. Our take is that the UK government considers that the robust evaluation of digital security flaws in AI models, as well as testing for other safety and security issues, is a key component of responsible AI deployment.

This approach has a somewhat different focus to that of the EU. The overall objective of the EU’s AI Act is to protect EU citizens’ fundamental rights and values. AI systems are categorised under the Act based on the risk they present to users’ health, safety and fundamental rights. The UK’s AI Security Institute is testing AI systems before they are released publicly and collaborating with major AI companies to improve the safety and security of their AI systems. Ultimately, the focus of both approaches is to keep the public safe but the EU’s approach is more closely grounded in protecting EU citizens’ fundamental rights and values.

Lawyers will be wondering what Liz Kendall means by ‘best practice’. To what extent will the best practice for evaluating AI models be mandated? Will there be a legal framework of sorts to ensure that the best practice is adhered to? Or will it be entirely voluntary? If so, how effective would it be? Based on the incumbent government’s past decisions, our expectation is that the guidance will be voluntary but we are watching closely. Perhaps things will become clearer in July when the institutes meet.

This article first appeared on Lexology | Source