Governance and regulation

We asked respondents what, if anything, would make them more comfortable with the use of AI. Participants could select multiple options. Figure 10 presents a comparison between the 2022/23 and 2024/25 survey results, illustrating changes in public attitudes towards mechanisms that enhance comfort with AI technologies. 

The majority of the public (72%) indicated that laws and regulations would increase their comfort with AI technologies – an increase from 62% in 2022/23. The second most commonly selected mechanism was the ability to appeal AI-generated decisions (65%), highlighting a strong public desire for avenues of redress in AI-driven decision making.

As highlighted previously, there is some latent discomfort with AI-generated decisions in the absence of an appeal mechanism. To explore this further, we examined people’s comfort with AI-generated decisions that impact their lives, and the role that explainability of those decisions would play to allay their discomfort. 

Two-thirds of the UK public (63%) are not comfortable with AI systems making decisions that affect their lives. In particular, those with fewer digital skills and lower incomes are slightly more likely than the nationally representative sample to report discomfort with automated decision-making systems, at 69% and 68% respectively, with this difference being statistically significant. Figure 11 shows overall how comfortable people are with AI technologies being used to make decisions that affect their lives. 

This discomfort sits alongside a preference for explanations to accompany decisions that affect people. To understand how the public make trade-offs between explanations accompanying AI decisions and the accuracy of these decisions, we informed participants that: ​‘Many AI systems are used with the aim of making decisions faster and more accurately than is possible for a human. However, it may not always be possible to explain to a person how an AI system made a decision.’ They were then asked to consider a range of statements around automated decision making and trade-offs between accuracy and explanations to accompany those decisions.

The public have a strong preference for explanations over accuracy. 62% of people think an explanation should always accompany a decision, with 37% feeling humans, not computers, should be the ones making these decisions. Only 8% of people think accuracy is more important than providing an explanation when it comes to automated decision making by an AI system. These findings are consistent with the previous survey wave, suggesting little change in the last two years in preferences for explanations over greater accuracy in automated decision-making systems. Figure 12 shows the distribution of responses when considering trade-offs between accuracy and explanations.

People’s preferences for explainability over accuracy are different across age groups. Older people choose explainability and human involvement over accuracy to a greater extent than younger people. For those aged 18–34, ​‘sometimes an explanation should be given even if it reduces accuracy’ was the most popular response (Figure 12). In contrast, for those aged 55 and above, the most popular response was ​‘humans should always make the decisions and be able to explain them’. This difference is also consistent with findings from our previous survey wave.

We asked the public about their experiences of the following types of online harms that may have been AI-generated: financial fraud or scams, deepfake images or video clips, false information, and content that promotes violence, abuse or hate. Figure 13 shows self-reported personal exposure to these harms. 

On average, close to two-thirds of the UK public (67%) have experienced any form of harm a few times, while over a third (39%) have encountered any form of harm many times. Exposure to harm was highest for false information, with 61% of people having experienced this, followed by financial fraud (58%), deepfakes (58%), and content promoting violence, abuse or hate (39%). However, many individuals are unsure if the harms they encountered online were AI-generated, with at least 20% reporting this for each of the harms we surveyed. 

Exposure to these harms is associated with age. Individuals aged 18–24 were more likely than other age groups to report having experienced these harms, with 81% reporting exposure to false information and 85% having encountered deepfakes. In contrast, older age groups reported different patterns of exposure to these harms. Those aged 65–74 reported encountering financial frauds (57%) and false information (53%) more commonly than other AI-generated harms. 

Men were more likely to report encountering online harms that were potentially AI-generated than women and this is statistically significant across harms. This aligns with research on public exposure to AI-generated harms such as deepfakes.¹ However, it is important to note that lower self-reported exposure among women could be due to their adaptive behaviours, including limiting their online engagement with others or limiting their own online engagement such as sharing photos or opinions online.² Such proactive behaviours, which may be driven by heightened fears about becoming a target of online harms,² might successfully reduce exposure to certain types of online harms, including AI-generated harms. However, as described in the limitations section, we do not have sufficient data to carry out detailed analysis. 

We also asked the public about the extent to which they were concerned about the spread of harmful AI-generated content online. Figure 14 shows self-reported concern about AI-generated harms. An overwhelming majority (94%) of the UK public said that they were either very or somewhat concerned about the spread of such harms online. Given these safety risks and concerns, it is important to understand public expectations around the regulation and governance of AI. We asked respondents who they think should be responsible for the safe use of AI and what specific powers they should have.

We asked the public about their expectations around the involvement of different stakeholders in AI safety. When asked who they think should be most responsible for ensuring AI is used safely, the majority of the UK public think an independent regulator (58%) and the companies developing AI technologies (58%) should be. Figure 15 shows expectations of responsibility. 

However, within this overall view, the preference for an independent regulator to be most responsible for ensuring AI is used safely increases with age, while preference for the companies developing AI decreases with age. Those aged 18–44 show a preference for companies over regulators, while those over the age of 55 have a preference for regulators over companies. This pattern of preference was similar to that found in our previous survey, suggesting age may continue to relate to expectations of, and potentially trust in, different organisations and institutions involved in AI development and deployment. Figure 16 shows responses across age groups to this question.

We also noted some regional variation in stakeholder preferences. For example, Northern Ireland shows a stronger preference for an independent oversight committee with citizen involvement (48%) compared to other nations. Further details can be found in Appendix section ​‘The four UK nations may have differing preferences around AI governance’. However, due to small sample sizes by nation, we cannot investigate regional differences in more depth in this report. 

We asked the public how important it was to them that the government or independent regulators have a series of specific powers around the use of AI. These were the power to:

• stop the use of an AI product if it causes harm

• actively monitor the risks posed by AI systems

• develop safety standards on AI use

• access information about the safety of AI systems from developers.

These powers were chosen because they relate to live issues around AI regulation, which may be pertinent for the UK government’s potential development of a future AI bill. Currently, there are no legal requirements for AI developers or independent regulators to regularly test or monitor upstream AI foundation models for safety risks, or statutory powers that allow regulators to restrict the sale of AI products or services outside of narrow sectoral regulation such as that for medical devices.

The public feel strongly about the government or independent regulators, rather than private companies alone, having a suite of powers related to the use of AI. 88% of people believe it is important that the government or regulators – and not just private companies – have the power to stop the use of an AI product if it is deemed to pose a risk of serious harm to the public. Figure 17 shows responses across the powers. The low prevalence of ​‘don’t know / prefer not to say’ responses suggests that these preferences are widely held.

When looking at the range of specific concerns people chose in relation to each AI technology, we found that more people report feeling concerned about the safety and security of their personal information across most uses of AI this year than in our previous survey wave. For example, this year 56% of people are concerned about facial recognition technologies in policing gathering and sharing their personal information with third parties. This figure was 38% in 2022/23. Similarly, 33% of people are concerned about the safety and security of their personal information in relation to AI technologies assessing welfare eligibility, with this figure at only 19% in our previous survey. Table 7 shows the prevalence of personal information concerns for uses of AI in 2022/23 and 2024/25.³

We asked the public about their views on data sharing in the public sector. Most (83%) are concerned by the idea of public sector bodies sharing data about them with private companies to train AI systems. This concern appears to be strongly held with very low proportions of the population feeling they did not know how concerned they were, if at all (3%), as shown in Figure 18. These concerns are important in the UK context where there have been explorations of anonymised NHS data sharing with private companies.⁴ 

We also asked the public about the extent to which they felt their views and values are represented in current decisions being made about AI and how it affects their lives. Half of the UK public (50%) said that they do not feel represented in this decision making, while just over a quarter (27%) said they do. Not feeling represented increases with age, with 57% of people aged 65–74 feeling unrepresented, compared to 40% of people aged 18–24.

Figure 19 shows the nationally representative distribution of responses to this question alongside breakdowns by age. 

Image credit: Alan Warburton / https://betterimagesofai.org / Image by BBC