Great (public) expectations: New research shows the growing disconnect between the public and government on AI regulation

The UK government and the EU are opting for a lighter-touch approach to AI regulation than had once been the plan – yet research by the Ada Lovelace Institute and others shows that the public are in favour of stricter binding controls on AI. In this opinion piece, Imogen Parker of the Ada Lovelace Institute argues that ensuring AI is ‘safe before sale’ is imperative to building the public’s confidence and enabling deployment

One of the UK Labour Party’s manifesto commitments, ahead of last year’s general election, was to “ensure the safe development and use of AI models by introducing binding regulation on the handful of companies developing the most powerful AI models”. 

Eighteen months after being elected, and despite generative AI becoming more powerful and increasingly used, the Labour government is yet to propose a bill on AI. And in last week’s Science and Technology Parliamentary Committee hearing, Liz Kendall, secretary of state for science, innovation and technology, ruled out a “big all-encompassing bill”, favouring the policy strategy of identifying specific areas requiring regulation.

This troubling trend toward deregulation extends beyond the UK. The EU – long acknowledged as setting a relatively high bar in data and AI regulation – is showing an inclination to roll back existing (and hard-won) legal protections around data, which powers AI models. If the EU’s ‘simplification’ package of changes in the Data and AI Omnibus is passed (affecting GDPR, the ePrivacy Directive and the AI Act), we will witness the greatest reversal of digital rights in a generation.

The irony? The disinclination to regulate comes at a moment when the public have never wanted regulation more.

Misaligned

In our 2024 survey, 72% of respondents said that laws and regulation would increase their comfort with AI – a full 10 percentage points higher than responses to the same question in 2022/3.

To drill into the details of that headline, we ran another nationally representative polling in late 2025 and surveyed the opinions of 1,928 adults living in the UK, focusing only on the regulation of AI. The findings underscore a severe misalignment between the government’s priorities around AI and the public’s.

For example, the public overwhelmingly stated that:

1. Safety should come before speed. When faced with trade-offs, 89% of respondents agreed that ‘AI products and services should not be rolled out until they are proved to be safe, even if this slows things down’.
2. Economic arguments alone aren’t enough to justify roll out. Almost three-quarters agreed that ‘some uses of AI should be restricted or banned on ethical or social grounds, even if they bring economic benefit’. 
3. Sovereignty matters. Less than a quarter of people disagreed that the ‘UK government should prioritise supporting UK AI companies, even if that means not using more powerful AI products and services from other countries’. The public also prioritises safety and accountability over international competition – only 38% believe that ‘it is important to keep up with other countries on AI, even if this means lighter rules’.

Disenfranchisement and distrust

The groups that perceived themselves as most likely to be negatively affected by AI – such as those without higher education qualifications, lower digital confidence or older people – are also those who feel least able to shape how AI is governed.

This feeling of exclusion is intensified by the public’s perception that the government prioritises the needs of technology companies over public interest: 84% of people are concerned about the government ‘putting the needs of the technology sector ahead of the public when regulating AI’.

Yet over half of the public (51%) say they ‘do not trust large technology companies, including social media companies, to act in the public interest’. This finding aligns with our previous public attitudes research, as well as the government’s own studies.

Read more: License to build: understanding what people think about public sector use of AI

Regulation with teeth

Unsurprisingly therefore, the vast majority of respondents support the independent regulation of AI: 89% say it’s important that AI is regulated independently, with only 1% stating it is ‘not at all’ important.

There is majority support for government or regulators to have the power to remove harmful tools from public access (85%) and halt their use and distribution if they are found to cause harm (89%), and for mandatory safety testing (82%).

The public also support transparency across a range of factors, including the societal costs of AI, its impact on communities and the environment (85%), and the economic costs, such as energy and resource use (82%).

Read more: Australia urged to set ‘rules of the game’ to harness the productivity potential of AI

Adoption above safety?

While the government has slowed down on the essential safeguards needed for frontier AI, it has focused on enabling approaches to adoption. The Department for Science, Innovation and Technology is prioritising the use of AI sandboxes – controlled use cases, in which businesses can test innovative products with regulatory exemptions.

Yet, as public trust in government and companies is low, the need for independent oversight and governance mechanisms are essential to deliver on adoption and confidence in AI.

This is not just a matter of the attitudes of the general public. In our research with public sector organisations, we repeatedly hear concerns and questions about regulation, ethics and legality of AI – particularly concerning newer frontier models. Just last week, the Royal College of General Practitioners and Nuffield Trust’s study into AI and GP practice highlighted that, despite numerous areas of optimism for AI use by GPs, the large majority perceived the lack of regulatory oversight to be a barrier to adoption.

Looking at another type of AI – biometrics – the government is finally consulting on a new legal framework for law enforcement use because the lack of it has so far inhibited widespread police adoption: many police forces have lacked the confidence to use newer biometric technology without proper guardrails.

AI adoption without adequate safeguards leaves risk flowing from companies to the public sector, and even down to local services and frontline staff. But it’s not just a public sector problem: AI minister Kanishka Narayan acknowledged “for many business leaders… ethical concerns remain a major barrier to adoption”.  

Read more: UK government unveils AI regulation blueprint to spur innovation across the economy

Calling time on AI builders-vs-blockers dichotomy

At the start of the year, prime minister Kier Starmer announced the UK would “go our own way on [regulation]. We will test and understand AI before we regulate it… but at the same time, we’ll offer the political stability that business needs. And, of course, our British values… rooted in democracy and the rule of law”.

That approach simply isn’t delivering clarity or stability. A lack of regulation is in fact blocking the ability to meaningfully test and understand AI, as user organisations are unable to get under the bonnet of the models they are procuring. Far from guaranteeing stability, the legal issues arising from foundation models are so unclear and unpredictable that major insurers are seeking to exclude AI risks from corporate policies.

Safe, effective, well-targeted AI technologies could improve public services – and some studies see real (although uncertain) opportunities for improving productivity.

Ensuring AI is safe before sale, and costs are counted and factored in, is a faster route to promote public confidence in adoption and, in turn, support for use. Without robust regulation, the government’s big bet on AI risks blocking deployment and stalling diffusion.

Read more: Transcribing trust: is transcription the use case that shows AI’s transformative power?