AI-Driven Disinformation as a New Security Challenge: Threats and Governance

1.Introduction

World Economic Forum (2024) identifies misinformation and disinformation as the most serious short term global risk. This is because the rapid development of generative AI has greatly increased the ability to produce and spread false information while also reducing the cost of doing so. The large scale spread of AI generated fake images and videos on the internet, together with the fake robocall incident during the 2024 United States election, shows that AI has already changed the form and scope of disinformation, and has made it one of the most important new security challenges.

This paper focus on disinformation in the context of the rapid development of AI. It uses specific case studies to show why it may become a future critical security threat, and discusses possible policy responses.

This paper is divided into three parts. The first part provides background information, including the definition, development process, and main characteristics of disinformation. The second part analyses the threats that disinformation may create in the context of rapid AI development. These threats include causing incorrect decision making, undermining social trust, damage democratic politics, expanding the range of threat actors, and creating difficulties for legal attribution. The final part examines existing policies and proposes further policy responses to the threats created by disinformation.

2.Background

2.1 Definition and Development of Disinformation

Different scholars have different definitions of disinformation. Rid (2020) argues that disinformation is a form of active measures carried out by state intelligence agencies, political organisations, and large bureaucratic systems. Its purpose is to manipulate information, create suspicion, and increase division in order to weaken rival societies. Freelon and Wells (2020) argue that disinformation has three core characteristics, deception, potential for harm, and intent to harm. Aïmeur et al. (2023) place greater emphasis on the differences between disinformation, misinformation, and malinformation. Misinformation refers to false information that is spread without the intention to mislead. Malinformation refers to true information that is spread in order to cause harm. Disinformation refers to false information that is deliberately created and spread.

Overall, most definitions of disinformation emphasise two main characteristics, false content and deliberate intention. Therefore, in this paper, disinformation refer to false information that is deliberately created or spread in order to influence public perception, create confusion, and manipulate public opinion. Due to the dependence of modern society on the internet, disinformation does not only affect politics and the military, but also influences the economy, society, culture, and many other areas.

Modern disinformation developed between the 1920s and 1940s. During this period, the Soviet secret police created fake anti communist organisations to mislead their enemies. After the beginning of the Cold War, both the United States and the Soviet Union started to systematically conduct political warfare and forge documents. During the 1970s and 1980s, the use of these methods expanded across the world (Rid, 2020). This shows that disinformation was not created by the internet or AI. Instead, contemporary technology has allowed it to reach an unprecedented scale and speed.

2.2 Main Characteristics of Disinformation

This paper argues that in the context of rapid AI development, disinformation as a security threat has four main characteristics, low production costs, expanded influence, increased deceptiveness, and asymmetric offensive and defensive capabilities.

Low production costs. In the context of the rapid development of generative AI, the cost of producing disinformation has been greatly reduced. By using generative AI, individuals can create large amounts of realistic false information in a short period of time, including both text and video content. This has significantly reduced economic and time costs. Experiment show that models such as GPT 3.5 and GPT 4 are already capable of generating convincing disinformation. AI models also tend to cooperate with users, and safety restrictions can often be bypassed through changes in wording and prompts (Vinay et al., 2024). In the past, producing disinformation required professional skills such as writing, programming, and video editing. Generative AI has lowered these technical barriers, while also reducing labour and knowledge costs (Hassoun et al., 2024).

Expanded influence. The rapid development of the internet has allowed disinformation to spread quickly across the world. With continuing technological development, the influence of disinformation has expanded even further. There are also large numbers of social bots on social media platforms. After combining with AI, these programs can use automated reposting and commenting to spread disinformation on a large scale (Shao et al., 2018). By analysing impressions related to COVID 19 and climate change on Twitter, Corsi (2024) found that low credibility tweets received greater exposure. In order to attract user, AI recommender systems actively push content based on user behaviour, which further increases the influence of disinformation.

Increased deceptiveness. With the development of generative AI technology, it has become much more difficult to identify manufactured disinformation. In terms of text, creators can train AI to learn specific writing styles in order to generate more realistic content. Through large amounts of training data, AI can also create videos that appear to show real people speaking, and can even fake speeches by Barack Obama (Chesney & Citron, 2019).

Asymmetric offensive and defensive capabilities. This asymmetry is first reflected in costs. In cyberwar, the defender must protect the entire system, while the attacker only needs to find one vulnerability (Libicki, 2009). In the case of disinformation, those resisting disinformation need to conduct long term and continuous content moderation, while those producing disinformation can achieve partial success as long as one piece of false information enters the public sphere. At the same time, false news may spread more easily than true information because it is more likely to create surprise, fear, and disgust (Vosoughi et al., 2018).

This asymmetry is also reflected in technology. Although both sides use AI technology, producing disinformation is often easier than detecting it. Aïmeur et al. (2023) argue that AI is also used to detect false information, but due to problems such as the inability of AI to fully understand context, the constant evolution of false information, and slow speed of fact checking, the ability of AI to generate false information has already exceeded its ability to identify it.

3.Major Threats of Disinformation

Based on the characteristics of contemporary disinformation discussed above, it can be seen that AI has already changed the form and scope of disinformation. It also creates five major threats, causing incorrect decision making, undermining social trust, damage democratic politics, expanding the range of threat actors, and creating difficulties for legal attribution.

3.1Causing Incorrect Decision Making

During the Russia Ukraine war, a video appeared on a Ukrainian news website showing Ukrainian President Volodymyr Zelenskyy calling on the military to surrender. Although the quality of the video was poor and it was quickly identified as deepfake generated false content (Twomey et al., 2023), it still showed that AI related disinformation had already been used in the context of war. During the 2025 clash between India and Pakistan, AI generated surrender statements and fabricated military operations also appeared online (Mah, 2026). This shows that AI generated disinformation is increasingly being used to fake orders from military leaders.

Research shows that under conditions of limited time and high pressure, people’s ability to distinguish between true and false information declines significantly, and they are also more likely to believe content that matches their own political views (Sultan et al., 2022). In high pressure situations such as ongoing international conflicts or moments that require rapid decision making, this type of disinformation may cause decision makers to make incorrect judgments or weaken their ability to think rationally, which may ultimately affect the international situation.

Disinformation does not only affect the decisions of leaders and policy makers, but also influence the judgment of news media and citizens. For news media organisations, manufactured disinformation can attract media attention and force journalists to spend time and resources verifying and reporting this content. Even if the information is later proven to be false, the process itself still consumes large amounts of resources and affects the original media agenda, causing genuinely important issues to be ignored or delayed.

For citizens, as discussed above, people are more likely to believe content that matches their own views and emotions. In order to attract attention, disinformation is often more focused on novelty and emotional reaction, with the aim of creating surprise, fear, or outrage (Vosoughi et al., 2018). The emotions created by disinformation may affect people’s work arrangements and investment decisions, which can lead to economic losses. At the same time, they can also influence public views on international affairs and domestic politics, which may intensify social divisions. Even if the information is eventually proven to be false, the uncertainty created by disinformation is itself a form of weapon and security threat (Byman et al., 2023).

3.2 Undermining Social Trust

The damage caused by disinformation to social trust is reflected in several aspects. First, the influence of disinformation on society is long term. Lewandowsky et al. (2012) proposed the continued influence effect. Even after false information has been corrected, people may still use that false information in their reasoning and judgment because of repeated exposure and the psychological tendency to fill logical gaps. This show that the influence of disinformation does not disappear after clarification. Instead, it may remain in people’s memories and continue to shape their understanding and judgment over a long period of time, creating division and misunderstanding in communication.

More importantly, disinformation often imitates real news and media content, creating an environment in which it becomes difficult to distinguish truth from falsehood. This may cause the public to lose trust in news and facts. Gallup data shows that public trust in whether news media report the news accurately and fairly has continued to decline in the United States since the 1970s, and reached a historical low point in 2016 (Freelon & Wells, 2020). The public has increasingly started to distrust all information and instead make judgments based on personal relationships, emotions, and political positions, which contributes to the formation of echo chambers (Ireton & Posetti, 2018). The lack of shared understanding of basic facts makes communication more difficult, gradually undermines social trust, and increases polarization.

Finally, Chesney and Citron (2019) proposed the concept of the “liar’s dividend”. As deepfake technology becomes more realistic and the public becomes aware that audio and video content can be manipulated, people who are actually lying may use this uncertainty to avoid responsibility. For example, during the Russia Ukraine war, a video showing Vladimir Putin appearing to move his hand through a microphone was actually caused by a video artifact, but many people believed that the video had been created using deepfake technology (Twomey et al., 2023). This show that the more cautious the public becomes about disinformation, the more likely they are to doubt genuine information as well, which weakens overall social trust.

3.3 Damage Democratic Politics

Disinformation does not have physical forms such as bombs or tanks, and it often appears in information warfare rather than traditional warfare. Information warfare is different from traditional warfare because it relies on technology, has lower costs, and creates greater political influence. It can cause enemies to make incorrect judgments, respond slowly, or fall into internal disorder (Bellamy, 2001). More importantly, information warfare does not have a clear beginning or ending, but instead exists within a difficult to identify grey area. As a hidden, low intensity, and continuous form of attack, disinformation is blurring the boundary between war and peace, and politic is becoming a new target of attack.

During the 2024 United States election, AI was used to fake the voice of Joe Biden in robocalls that encouraged voters not to vote. Similarly, during elections in India, AI generated videos of celebrities and deceased political figures were widely used for mobilisation and political campaigning(Lopes, 2025). This shows that disinformation can play two different roles in elections, suppressing voter participation and mobilising voters.

Although this type of disinformation has not yet directly changed the result of an election, some voters have already developed fears that disinformation may manipulate elections. This fear of manipulation is itself harmful to the electoral system.

The political influence of disinformation is not limited to elections. It may also damage democratic systems themselves. Pawelec (2022) argues that modern democracy depends on the ability of the public to engage in deliberation and political decision making based on accurate information, while AI generated disinformation technologies such as deepfake weaken this foundation. As discussed above, disinformation not only misleads citizens’ judgment, but may also cause people to distrust all information. When citizens cannot determine what information is true, rational political discussion and political judgment become much more difficult.

Disinformation may also be used to suppress freedom of speech and public participation. In 2018, a fake porn video created using the face of Indian investigative journalist Rana Ayyub spread online. The purpose was to threaten and intimidate her in order to stop her from continuing to report on religious violence, human rights issues, and the Indian government (Pawelec, 2022). This kind of attack may cause minorities, vulnerable group, and journalists to reduce public expression because of fear, which limits the legitimate pursuit of rights and equality in democratic politics.

3.4 Expanding The Range of Threat Actors

Historically, disinformation was mainly organised by state intelligence agencies. States could attack their opponents through forged documents, covert propaganda, front organisations, and leak operations (Rid, 2020). This form of information warfare, which was mainly carried out by states, required professional personnel, long term planning, and large amounts of resources. With the development of AI technology, the cost of producing disinformation has been greatly reduced, allowing a growing number of actors to participate in the creation and spread of disinformation.

Due to the spread of AI generation technology and the large amount of open source code available online, technology and financial resources are no longer the main barriers. Instead, the ability to produce and spread disinformation increasingly depends on computing power and CPUs. This allows not only economically and technologically weaker states to gain information manipulation capabilities that were previously mainly controlled by major powers, but also enables non state actors to participate, including international organisations, companies, social groups, and even terrorist organisations.

Taking network communities as an example, in 2017, an anonymous user began posting mysterious messages on 4chan that could be interpreted in different ways, gradually forming a conspiracy theory network community. Baća (2024) argues that this was an epistemic community of the unreal. Members were not simply passive receivers of propaganda, but were actively producing meaning through interaction, creation, and collective organisation. AI technology has made information more difficult to distinguish from reality and has also provided these online communities with more materials and wider space for interpretation. As a result, increasing numbers of people are able to incorporate their own national cultural and social conflicts into this interpretive system, allowing disinformation and related feelings of fear and anxiety to spread globally.

Ordinary citizens may also become part of the expansion of these threats. Golovchenko et al. (2018) studied the shooting down of Malaysia Airlines Flight MH17 in eastern Ukraine in 2014. Among the 50 most central accounts spreading misleading false information, 39 were ordinary citizen accounts rather than government or official media accounts. This show that in the age of social media, ordinary citizens are no longer only receivers of information, but also actors with significant influence, both in spreading and resisting disinformation. This means that when responding to disinformation, attention cannot only focus on state actors or professional organisations. It is also necessary to deal with large amounts of content produced by ordinary citizens, which greatly increases the workload, cost, and difficulty of defending against disinformation.

3.5 Creating Difficulties for Legal Attribution

Disinformation has strong concealment and anonymity for two main reasons, technological methods and organisational methods. In terms of technology, the creators and spreaders of disinformation can hide their identities online through methods such as proxy servers, virtual networks, and disguised Bluetooth devices, making it difficult for investigators to trace the real source of information. At the same time, with continuing technological development, large amounts of AI generated content are spreading virally across the internet. Researchers have found that AI generated content is creating a new pattern of information dissemination(Chrysidis et al., 2026). In both image and video forms, it is more likely to spread virally than other types of false information. Large amounts of highly similar content can also create confusion and reduce the efficiency and accuracy of investigations.

In terms of organisational methods, state and organisations do not need to directly carry out cyber operations themselves. Instead, they can rely on proxies, contractors, or financially supported hacker groups to achieve their goals. Taking the Syrian Electronic Army as an example, the hacker group hacked the Twitter account of the Associated Press in 2013 and spread disinformation claiming that there had been an explosion at the White House and that the President of the United States had been injured, which caused a sudden fall in the United States stock market. In the same year, the group also hijacked the domains of The New York Times and Twitter. Through analysis of domain registration records, organisational structure, and attack targets, Al Rawi (2014) argues that these operations were not simply independent hacker activities, but were closely connected to the government of Bashar al-Assad. This type of indirect cyber operation allows governments to spread false information and disrupt social stability while also claiming that the actions were carried out independently by non state actors, thereby avoiding direct attribution under international law.

On the other hand, the effects created by disinformation are different from traditional physical attacks because their influence is often indirect and long term. International law usually deals with physical causation, while disinformation involves mental causation. The complex chain of causation makes its effects difficult to measure accurately (Lahmann, 2022). Cyber operations involving disinformation often make it difficult to identify a clear attributable actor and to estimate the impact of the operation, which creates major challenges for the application of international law to these types of attacks.

This shows that difficulties in attribution not only make it harder to identify the real attackers and create challenges for defence and content moderation, but also prevent relevant institutions from making accurate responses in a timely manner. This leaves more space for the spread of disinformation and directly threatens economic and social stability. At the same time, this governance vacuum may increase suspicion and distrust between states, and may even escalate conflicts because of misjudgment and overreaction.

4.Policy Responses and Recommendations

4.1 Existing Policies and Governance Frameworks

While AI is changing the world, people have also gradually become aware of the threats and risks created by AI generated disinformation. The United Nations, the European Union, national governments, and companies have all started to respond to this security threat.

The United Nations discussed international AI governance in the 2024 report Governing AI for Humanity. The report shows that one of the main concerns among AI experts is the damage that misinformation and disinformation may cause to the public information environment (United Nations, 2024). The report calls for international cooperation, the establishment of global cooperation mechanisms, coordination of global AI standards, and support for AI development in developing countries. In the Global Digital Compact, the United Nations (2024) also encourages states to strengthen media literacy education, support independent media, and improve fact based information dissemination.

Although the United Nations has continued to pay attention to the risks created by disinformation, most of its responses are based on soft law measures such as recommendations and guidelines, rather than binding international law. International governance of disinformation still depends heavily on the voluntary cooperation of states. Because countries have different national conditions and different understanding of disinformation, global governance still lacks unified and enforceable standards.

The European Union is currently one of the strictest actors in the regulation of generative AI and disinformation. The Digital Services Act and the EU AI Act are the main regulations used by the European Union for information governance. The Digital Services Act requires platforms to assess and reduce disinformation risks and improve algorithmic transparency. The EU AI Act strengthens the regulation of AI technology and requires AI generated content to be labelled. These two policies show that the governance approach of the European Union places strong emphasis on algorithms. This reliance on techno solutionism suggests that the European Union is more focused on responding to current crises in disinformation governance, while lacking a broader framework and long term vision (Wijermars & Makhortykh, 2022).

There are significant differences in how states govern disinformation. Some countries adopt a government dominated model, such as China and Nigeria, while others use a platform dominated governance model, such as the United States and Sweden (Ma et al., 2025). Governments often strengthen regulation over online platforms, increase content moderation, and require AI generated content to be labelled. At the same time, this creates tensions between disinformation governance and freedom of speech. The United Nations Human Rights Council (2021) criticised many states for using vague and overly broad laws to combat so called fake news, which may violate international human rights law. At the same time, giving social media platforms the power to moderate content may also create problems such as over removal and lack of due process.

Large technology companies and social media platforms have also become involved in the governance of disinformation. Companies such as Meta, Google, YouTube, and TikTok have established their own content moderation systems. For example, YouTube uses information panels, fact check panels, and authoritative source links to reduce the spread of disinformation (YouTube, n.d.). Current governance strategies rely heavily on AI technology and algorithms. At the same time, the lag between technological development and regulation makes it difficult for platforms to identify and remove all forms of disinformation. More importantly, platform moderation standards and enforcement are also influenced by commercial interests. Social media algorithms tend to promote emotional and conflict driven content in order to increase engagement.

Overall, although different levels of governance have already developed measures to respond to disinformation, there are still several major problems, including the lack of unified standards, the inability of regulation to keep pace with technological development, conflicts with freedom of expression, and the neglect of power structures.

4.2 Policy Recommendations

Establish international norms.Today, disinformation is not only a problem of domestic information governance, but also a tool of state competition. When disinformation governance conflicts with national interests and military security, morality and domestic laws are often insufficient to restrain government behaviour (Emery Xu et al., 2025). In addition, because disinformation is now widely spread through digital networks, its threats can ignore geographical boundaries and spread globally across states. Therefore, the international community needs to establish a more unified global governance framework.

First, the international community should establish an international definition of disinformation, especially AI generated disinformation, clarify its impact, and analyse its possible future development. This would help promote international cooperation by building global consensus.

Second, international cooperation and supervision mechanisms should be strengthened. Specialised international institutions could monitor transnational bot networks, focus on high risk disinformation content, and investigate information warfare between states. This would improve the coordination and standardisation of global disinformation governance and reduce conflicts between states.

Finally, the governance of disinformation also requires the reinterpretation of relevant international law in order to incorporate disinformation into the existing legal framework. For example, when evaluating the effects of disinformation, international law should not only focus on physical damage, but should also pay greater attention to non physical harm and the long term development of situations. It is also possible to draw on the approach of the Tallinn Manual 2.0 regarding due diligence in cyber operations by requiring states not to knowingly allow their territory to be used for the large scale production and spread of disinformation. This may help make progress on the problem of attribution.

Strengthen moderation and oversight.With the development of technology, social media platform are becoming increasingly dependent on AI for content moderation and recommendation. As a result, the transparency of AI models and algorithms used by platforms has become more important for platform oversight. Burnat and Davidson (2026) argue that during the 2023 Israel Gaza conflict, Meta was accused of systematically suppressing Palestinian content. At the same time, because Meta restricted access to platform data, society was unable to effectively verify and monitor these events. The lack of oversight over algorithms and AI models makes it difficult for the public to know what types of content are being promoted and amplified. Therefore, this paper argues that under the existing background of the Digital Services Act, more countries should learn from the policies and experience of the European Union and introduce laws that improve platform algorithm transparency, so that the operation of online platforms can face stricter supervision. At the same time, countries that already have relevant laws and policies should strengthen the quality of enforcement.

In relation to platform algorithms themselves, some researchers argue that harmful information may already have caused damage before it is removed, because the spread of the content has already taken place. Therefore, platforms should pay greater attention to “content likely to go viral in the future” (Merrill & Oremus, 2025). When responding to disinformation, platforms should also improve their moderation processes and methods. They should not only examine the content itself, but also focus on its influence and potential reach. Platforms should prioritise disinformation that is likely to spread on a large scale in order to reduce exposure as much as possible.

Encourage the participation of more actors. The governance of disinformation is not only a technical issue, but also a political issue. As discussed above, state centred governance systems have raised concerns about the protection of freedom of expression and the neglect of power structures behind rule making.

NGOs can provide external oversight over governments and platforms, resist disinformation, and at the same time protect human rights and the rights of vulnerable groups. Research institutions can also provide professional technical support and improve the professionalism of governance. Civil Society Organizations can play an important role in governance as well. CSOs can cooperate with technology companies to trace the sources of disinformation and establish databases to help the public identify whether information is trustworthy. This can not only reduce the influence of disinformation, but more importantly, CSOs can help develop public critical thinking and information literacy during this process (Bejma & Kao, 2025).

Governments and platforms should strengthen their ability to identify AI generated disinformation. Platforms can provide more accessible information verification tools, while governments have the responsibility to organise more educational and public awareness activities and respond quickly when disinformation appears.

5.Conclusion

This paper focuses on disinformation in the context of the rapid development of AI and discusses the issue from two perspectives, the threats it may create and possible policy responses.

AI is changing the form and scope of disinformation. Due to it four main characteristics, low production costs, expanded influence, increased deceptiveness, and asymmetric offensive and defensive capabilities, this paper argues that disinformation will create five major threats in the present and future, causing incorrect decision making, undermining social trust, damaging democratic politics, expanding the range of threat actors, and creating difficulties for legal attribution. Although the international community has recognised the dangers of disinformation and some governments have already introduced policies to respond to this security threat, these policies still contain several weaknesses, including the lack of unified standards, the inability to keep pace with technological development, conflicts with freedom of expression, and the neglect of power structures. In response to these problems, this paper argues that addressing the threat of disinformation in the context of large scale AI application requires the establishment of international norms, stronger moderation and oversight, and the encouragement of the participation of more actors.

Reference list

Al-Rawi, A. K. (2014). Cyber warriors in the Middle East: The case of the Syrian Electronic Army. Public Relations Review, 40(3), 420–428.

Aïmeur, E., Amri, S., & Brassard, G. (2023). Fake news, disinformation and misinformation in social media: A review. Social Network Analysis and Mining, 13(1), 30.

Bejma, A., & Kao, Y.-J. (2025). Resilience to online disinformation: The role of civil society organizations. Polish Political Science Yearbook, 54(3), 89–104.

Bellamy, C. (2001). What is information warfare? In R. Matthews & J. Treddenick (Eds.), Managing the revolution in military affairs (pp. 56–75). Palgrave Macmillan.

Burnat, F. A. D., & Davidson, B. I. (2026). The accountability paradox: How platform API restrictions undermine AI transparency mandates. arXiv.

Byman, D. L., Gao, C., Meserole, C., & Subrahmanian, V. S. (2023). Deepfakes and international conflict. Brookings Institution.

Chesney, B., & Citron, D. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. California Law Review, 107(6), 1753–1820.

Chrysidis, Z., Papadopoulos, S.-I., & Papadopoulos, S. (2026). The synthetic media shift: Tracking the rise, virality, and detectability of AI-generated multimodal misinformation [Preprint]. arXiv.

Corsi, G. (2024). Evaluating Twitter’s algorithmic amplification of low-credibility content: An observational study. EPJ Data Science, 13(1), Article 18.

Emery-Xu, N., Jordan, R., & Trager, R. (2025). International governance of advancing artificial intelligence. AI & Society, 40(4), 3019–3044.

Freelon, D., & Wells, C. (2020). Disinformation as political communication. Political Communication, 37(2), 145–156.

Golovchenko, Y., Hartmann, M., & Adler-Nissen, R. (2018). State, media and civil society in the information warfare over Ukraine: Citizen curators of digital disinformation. International Affairs, 94(5), 975–994.

Hassoun, A., Abonizio, A., Osborn, K., Wu, C., & Goldberg, B. (2024). The influencer next door: How misinformation creators use GenAI [Preprint]. arXiv.

Ireton, C., & Posetti, J. (Eds.). (2018). Journalism, fake news & disinformation: Handbook for journalism education and training. UNESCO.

Lahmann, H. (2022). Infecting the mind: Establishing responsibility for transboundary disinformation. European Journal of International Law, 33(2), 411–440.

Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N., & Cook, J. (2012). Misinformation and its correction: Continued influence and successful debiasing. Psychological Science in the Public Interest, 13(3), 106–131.

Libicki, M. C. (2009). Cyberdeterrence and cyberwar. RAND Corporation.

Lopes, A. R. (2025). Programmed elections: The impact of artificial intelligence on the electoral process. Revista de Direito, Estado e Telecomunicacoes, 17(2), 319–342.

Mah, P. M. (2026). The role of deepfake, deception, and disinformation in conflict zones based on DL for NLP: A critical AI-era perspective. Comunicar, 34(84), 228–246.

Mariëlle Wijermars, & Mykola Makhortykh. (2022). Sociotechnical imaginaries of algorithmic governance in EU policy on online disinformation and FinTech. New Media & Society, 24(4), 942–963.

Merrill, J. B., & Oremus, W. (2025, April 29). Researchers call for new way of thinking about content moderation. The Washington Post.https://go.gale.com/ps/i.do?id=GALE%7CA837616368&v=2.1&it=r&u=usyd&p=AONE&aty=shibboleth

Pawelec, M. (2022). Deepfakes and democracy (theory): How synthetic audio-visual media for disinformation and hate speech threaten core democratic functions. Digital Society, 1(2), Article 19.

Rid, T. (2020). Active measures: The secret history of disinformation and political warfare (1st ed.). Farrar, Straus and Giroux.

Robert Gorwa, Reuben Binns, & Christian Katzenbach. (2020). Algorithmic content moderation: Technical and political challenges in the automation of platform governance. Big Data & Society, 7(1), 2053951719897945.

Rui Ma, Xueqing Wang, & Guo-Rui Yang. (2025). Fighting fake news in the age of generative AI: Strategic insights from multi-stakeholder interactions. Technological Forecasting and Social Change, 216, Article 124125.

Shao, C., Ciampaglia, G. L., Varol, O., Yang, K.-C., Flammini, A., & Menczer, F. (2018). The spread of low-credibility content by social bots. Nature Communications, 9(1), Article 4787.

Sultan, M., Tump, A. N., Geers, M., Lorenz-Spreen, P., Herzog, S. M., & Kurvers, R. H. J. M. (2022). Time pressure reduces misinformation discrimination ability but does not alter response bias. Scientific Reports, 12(1), Article 22416.

Twomey, J., Ching, D., Aylett, M. P., Quayle, M., Linehan, C., & Murphy, G. (2023). Do deepfake videos undermine our epistemic trust? A thematic analysis of tweets that discuss deepfakes in the Russian invasion of Ukraine. PLOS ONE, 18(10), e0291668.

United Nations. (2024). Global Digital Compact (A/79/L.2). United Nations.

United Nations. (2024). Governing AI for humanity: Final report. United Nations.

Schmitt, M. N. (Ed.). (2017). Tallinn manual 2.0 on the international law applicable to cyber operations. Cambridge University Press.

United Nations Human Rights Council. (2021). Disinformation and freedom of opinion and expression: Report of the Special Rapporteur on the promotion and protection of the right to freedom of opinion and expression, Irene Khan (A/HRC/47/25). United Nations.

Vinay, R., Spitale, G., Biller-Andorno, N., & Germani, F. (2024). Emotional manipulation through prompt engineering amplifies disinformation generation in AI large language models [Preprint]. arXiv.

Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146–1151.

World Economic Forum. (2024). The global risks report 2024 (19th ed.). World Economic Forum.

YouTube. (n.d.). Topical context in information panel. Google Help. https://support.google.com/youtube/answer/9004474?hl=en-EN