Written evidence submitted by The Alan Turing Institute (DIS0030)

Written evidence from The Alan Turing Institute, prepared by Samuel Stockwell, Research Associate, Centre for Emerging Technology and Security (CETaS), The Alan Turing Institute.

The Alan Turing Institute (hereafter ‘the Institute’) is the UK’s national institute for data science and artificial intelligence. Our teams working on Defence and National Security challenges conduct policy and technical research on emerging technologies. Within this context, understanding how malign actors are using  technology to erode our democratic values is a high priority. We therefore welcome the opportunity to provide evidence to this Committee investigation.

Who are the main state and non-state actors spreading disinformation?

1.1  With the emergence of new generative AI models since late 2022, restrictions on those seeking to spread misleading but realistic content have been significantly reduced, since users no longer need technical expertise to exploit these tools.[1]

1.2  As a result of these developments, there has been a widening of the disinformation risk landscape from foreign state-sponsored networks to both domestic extremist groups and individuals unaffiliated to larger organisations.[2]

1.3  From our analysis of recent elections throughout 2024, we have identified Russian-backed networks such as Doppelganger as some of the most prominent groups utilising generative AI tools to spread disinformation across multiple democracies. During the 2024 US presidential election, Chinese- and Iranian-linked networks also sought to influence voters and undermine societal cohesion via misleading AI-generated material.[3]

1.4  Alongside hostile state-sponsored organisations, domestic extremist groups such as the English Defence League have amplified harmful AI-generated content to stoke public unrest and violence. This included a deepfake audio clip targeting London Mayor Sadiq Khan during Armistice Day.[4]

1.5  Finally, individual users unaffiliated to any specific organisations are an increasing source of disinformation through generative AI models. During the 2024 UK general election, a small number of voters expressed their dissatisfaction towards certain political candidates with deepfake smear campaigns. This may not only damage the integrity of the information ecosystem due to the realism of such fake content, but can contribute towards a ‘chilling effect’ on the willingness of individuals to enter politics in the future over fears of reputational damage. Additionally, some users sought to promote harmful narratives calling for violence during the Southport riots through xenophobic deepfake images of Muslim men being chased.[5]

What channels and technologies are state and non-state actors using to spread disinformation?

2.1  Across elections in 2024, hostile state-sponsored actors were relatively sparing in their use of new generative AI tools as part of interference efforts. AI models primarily helped to enhance the realism, personalisation, and amplification of misleading content, as opposed to revolutionising these activities. This included: tailoring fake news stories on key issues such as Russia’s War in Ukraine to certain voter demographics; utilising bot accounts to spread conspiracy theories surrounding major campaign incidents, such as Donald Trump’s assassination attempt; and creating realistic deepfakes to undermine the reputation of unfavourable candidates, such as US Democrat Party politicians.[6]

2.2  In contrast to fears from government officials and experts, our analysis showed that hostile state-sponsored groups were more interested in undermining societal cohesion and deepening political polarisation with AI-generated disinformation during elections – as opposed to directly influencing results in specific constituencies. Indeed, Romania’s decision to annul its recent presidential vote over alleged Russian interference has led to controversy and division within the country, which may indirectly benefit the Kremlin by damaging domestic trust in Romania’s democratic system.[7]

2.3  During the 2024 US presidential election, we identified a separate but worrying trend where domestic conspiracy theorists and extremist groups were able to amplify political disinformation more effectively through AI-generated content than previously. In some cases, deepfake smear campaigns against female candidates by these networks also helped to spread sexist narratives.[8]

2.4  Instead of seeking to influence undecided or opposition voters, most of the AI-generated disinformation shared by domestic extremist groups were primarily reinforcing pre-existing partisan divides. For example, deepfakes targeting US Democrat Party candidate Kamala Harris were largely circulated within echo chambers of right-wing users. However, a deepening of such echo chambers can harden political polarisation in society and make it more challenging to preserve public trust in verified sources of information.[9]

What is the effectiveness of cross-departmental and inter-governmental coordination, alongside the private sector in countering state-sponsored disinformation?

3.1  Addressing state-sponsored disinformation requires an approach which recognises how, despite it being impossible to entirely prevent the proliferation of such content, friction points can be created to make these activities more difficult or expensive to conduct. Changing such incentives necessitates not only a cross-government response, but multi-sectoral collaboration with industry, civil society, and academia.

3.2  Interventions should focus on targeting different stages of the disinformation content life cycle, outlined in our recent CETaS framework. This helps to break down the problem space into a more manageable set of four action areas for different sectors: curtailing the generation, constraining the dissemination, and counteracting user engagement with deceptive online content; while also empowering society to expose and undermine online disinformation.[10]

24 February 2025