According to the report of the International Economic Forum 2024, disinformation is the most severe      global risk in the short term. This risk is driven, particularly      by the rapid development of generative artificial intelligence (AI), which      can create vast amounts of manipulative, "synthetic" content.

Artificial intelligence systems, particularly generative AI, which can create new data, such as images, text, or videos, are gaining increasing popularity and becoming a promising tool for various industries. However, it is possible that such technologies could be used for malicious purposes, especially to generate misleading      materials that harm society.

AI technologies as a Tool for Spreading Disinformation

One of the most common and well-known methods of creating disinformation using AI is the so-called deepfake –      technology that allows      the creation of realistic photos, videos, or audio materials (often referred to as “synthetic” content) by manipulating a person's appearance or voice. 

Deepfakes have found practical applications in several fields, particularly in the film industry, where the technology is used to recreate the voices or faces of actors and historical figures. The development of these technologies has given rise to tools such as DALL-E and Midjourney, which are available to the general public, making the creation      of materials using AI faster and easier. Although these tools are promising for professional and widespread use, they make it possible to create a significant amount of harmful disinformation content.

In particular, deepfake technology can be used in political propaganda and discrediting opponents. Back in 2017, researchers at the University of Washington (USA) developed a deepfake with former US President Barack Obama to demonstrate the capabilities of machine learning in creating realistic videos. In 2023, internet users and media platforms began circulating deepfake images of Donald Trump allegedly being arrested, as well as a fake video of Joe Biden, in which he says that men and women in their twenties will be selected to defend Ukraine.

Deepfakes have also started to be actively created following Russia's full-scale invasion of Ukraine. Perhaps  the most striking example is the deepfake video made in 2022 featuring the face of Volodymyr Zelensky, in which the president orders Ukrainian soldiers to lay down their arms and surrender. Although, at that time, the deceptive video was poorly made and quickly recognized as a deepfake, such technologies are capable of rapidly evolving and becoming more realistic.

Last year, a group of American researchers conducted an experiment to assess people's ability to distinguish authentic videos from deepfakes. The results showed that a significant percentage of Americans incorrectly identified the authenticity of videos – between 27% and 50% of respondents were unable to determine the truthfulness of the presented materials. This means that deepfake technologies are becoming good enough to mislead the viewer.

The use of AI technologies for disinformation purposes has also become possible due to the development of Large Language Models (LLM) such as ChatGPT, Bard, Gemini, etc. LLM tools can be used both to generate fake texts and news and to create online bots that mimic the personality of real people.

Another study conducted this year sought to determine whether propaganda created with GPT-3 was sufficiently persuasive to readers. The results show that propaganda news generated by artificial intelligence can be as convincing as original propaganda. At the same time, the persuasiveness of the content increased when the propaganda message was generated through collaboration between AI and humans. Researchers argue that with the emergence of more advanced models like GPT-4, AI-generated materials can amplify the volume of propaganda and complicate its detection.

During the investigation, the NewsGuard team found about a thousand unreliable websites generating news using artificial intelligence last year. According to experts, such sites operate with little or no human supervision and publish articles created mainly by bots. An example of this is an AI-generated article about the psychiatrist of the Israeli Prime Minister allegedly committing suicide. After its publication, the fake news began to spread across social networks and media platforms.

Generative AI has made it possible to create AI-based bots – automated accounts designed to mimic human behavior, often for malicious purposes. The New York Times' open-source chatbots have shown how vulnerable AI technologies can be to disinformation and propaganda. The capabilities of artificial intelligence allow not only the creation of propaganda messages but also the inclusion of typographical errors, emotionally charged words, and poorer coherence. Thus, bot comments can become more realistic, making them harder to identify as AI-generated.

The use of artificial intelligence to create and disseminate misinformation also raises significant concerns among the expert community.

Caycee Myers, a communications policy expert at Virginia Tech, a US research institute, says that the rise of AI-enabled disinformation has raised several practical and legal issues. "Easy access to AI means that disinformation, including fake news, is easier to create and spread, making it harder for lawmakers to track. Legal liability for fake content poses certain logistical challenges, as many of the people who create content cannot be identified or caught. Some of these individuals live outside the country where the disinformation content is published, complicating accountability," he notes.

Initiatives for AI Regulation Regarding Disinformation in the European Union and Ukraine

In the European Union, the use of deepfake technologies and other tools used for disinformation purposes is regulated by the EU AI Act and the Digital Services Act (DSA).

The recently adopted EU AI Act defines deepfakes as AI systems with a limited risk. This means that such technologies are subject to certain transparency obligations. According to Article 50 of the Act, providers of artificial intelligence, including deepfakes that manipulate images, videos, and audio materials or generate text, are required to indicate that the content was created using artificial intelligence. 

In 2022, the EU also adopted the Digital Services Act, which partially addresses the regulation of synthetic disinformation content in the online environment. In particular, online platform providers should pay attention to how their services can be used to spread misleading content, including through algorithmic systems, and assess content moderation measures to mitigate potential risks. 

Additionally, Article 35.1 of the Act requires online platforms to provide "prominent markings" of information, whether it is "a generated or manipulated image, audio or video that appreciably resembles existing persons, objects, places or other entities or events and falsely appears to a person to be authentic or truthful." 

It is important to note that the DSA provides a mechanism to engage Trusted Flaggers, non-governmental organizations that monitor online platforms for illegal content, including fakes created with the help of AI. The system is based on the voluntary cooperation of such organizations and online platforms to more effectively detect illegal content and ensure the protection of the rights of EU citizens.

Although EU legislation regulating the use of synthetic content imposes additional obligations on providers and platforms regarding the marking of AI-generated content, there is currently insufficient direct evidence to support the effectiveness of such marking. Moreover, existing marking methods may not be sufficient to mitigate the risks posed by the proliferation of synthetic content. 

For example, experts from the Massachusetts Institute of Technology (USA) note that AI-labeling such as visual labels, watermarks, audio labels, or text marks can be easily removed by manipulating images with editors. Also, attackers may intentionally fail to flag such content, undermining transparency efforts. There are also concerns that marking may reduce the information value of AI-generated material and undermine its credibility.

Since Ukraine is in the process of European integration, it is essential to gradually align national regulations with EU legislation, particularly in matters concerning the use of AI technologies. However, the Ukrainian government is already taking steps to combat AI-driven disinformation. In particular, the Ministry of Digital Transformation has proposed a range of general tools that should be implemented until future legislation is adopted. 

One such tool is the voluntary marking of AI systems, which involves developers publishing "clear and structured" information about their AI systems. The purpose of such marking is to increase transparency about AI products and cover the requirements set forth in the European legislation. Potentially, such marking could also apply to AI-generated content, which would also have an impact on combating disinformation. 

Additionally, the Ministry of Digital Transformation plans to release a series of sector-specific recommendations related to the application of AI technologies in various fields. The first document in Ukraine to provide such recommendations is the Guidelines on the Responsible Use of Artificial Intelligence in the Media. The Guidelines for the media aim to     ensure the responsible implementation of AI systems, adhere to professional standards, and specifically address the use of these technologies for processing, generating, and distributing content.

In the context of creating content using generative AI, media outlets are advised to disclose how and when such technologies were used. Additionally, if materials contain content generated by artificial intelligence, it is recommended to mark such content and avoid disseminating AI-generated materials intended to mislead or spread disinformation.

Following the example of the EU Digital Services Act, Ukraine will also implement a Trusted Flaggers mechanism to involve Ukrainian organizations in addressing user complaints regarding human rights violations related to the use of AI tools by leading platforms. In this way, the Ministry of Digital Transformation offers an additional tool for protecting human rights that does not require strict legislative frameworks. The potential field of activity for Trusted Flaggers might involve checking for misleading or disinformation content generated by AI.

Conclusions and Recommendations

AI technologies can have great potential in combating fakes but can also exacerbate disinformation risks. While AI as a tool is not inherently harmful, its accessibility creates more opportunities for malicious actors to produce synthetic fake content more quickly and on a larger scale. Moreover, AI-generated disinformation is becoming harder to detect, creating opportunities for manipulating public opinion and misleading society.

Currently, there are no universal tools for combating disinformation created using AI technologies. Most existing recommendations focus on educational activities related to media literacy and legislative changes. The EU is taking practical steps in this area, listing           countering disinformation among the goals of implementing the AI Act.

It is more challenging for Ukraine to acquire such effective tools, particularly because the EU is the key market for all online services and platforms. However, in the context of countering Russian aggression and considering the challenges that the state and society will face in the post-war period, certain steps must be taken right now.

Such measures may include:

1. Improving legislation regarding the combating of disinformation. This may require changes to current laws, particularly those related to media and information, as well as the adoption of legislation on AI and disinformation. The White Paper on AI regulation in Ukraine states that the relevant legislation should be developed after the end of the war. However, considering the importance of this issue for national security and the need to comply with relevant requirements and standards, it is advisable to explore the possibility of implementing sector-specific regulation based on the EU AI Act norms earlier. 
2. Improving marking mechanisms. As the White Paper states, the Ministry of Digital Transformation plans to introduce voluntary marking of AI systems, which may also apply to AI content. Although current methods of marking AI-generated content may be ineffective, labeling is still considered one of the beneficial and necessary tools ensuring protection against malicious AI content. In this respect, important steps include improving the accuracy of identifying generated content and defining the context and objectives of marking. The process of developing an effective marking mechanism will require a significant amount of testing but could have a positive impact on the fight against disinformation in the long run. And the introduction of such mechanisms could be part of the proposed changes to the legislation.
3. Strengthening cooperation with online platforms, particularly through the involvement of civil society. Establishing cooperation with online platforms can be a potentially effective informal measure to combat disinformation, especially at the stage of preparing Ukrainian legislation that would regulate the use of artificial intelligence. This is possible, in particular, through the Trusted Flaggers mechanism proposed by the Ministry of Digital Transformation. This initiative could not only enhance the experience of Ukrainian civil society organizations in identifying illegal content but also serve as a tool to protect society from the negative disinformation risks associated with the use of AI technologies.
4. Integrating AI tools to combat disinformation. Artificial intelligence is also a crucial tool in the fight against misinformation. The latest AI technologies make it possible to analyze patterns and context and detect AI-generated content. Today, many AI-based tools  help fact-checkers, media, and users detect fake news and prevent their      spread more quickly and efficiently. Such systems can be particularly helpful in the activities of Trusted Flaggers and government agencies responsible for information policy and security to assist with the identification of AI-generated content.    
5. Developing effective communication tools to combat the disinformation risks of AI. While many proposed approaches, such as educational programs on media literacy, may have limited effects, proper communication can help mitigate the risks associated with AI-generated disinformation. In particular, monitoring and prompt debunking of AI-generated fakes, indicating the methods of their creation, and communication with online platforms to remove them can significantly reduce the negative consequences. Such tools can be utilized in collaboration with the government, media, and civil society, which can significantly enhance the level of trust in them.