AI-Driven violence threatens journalists’ safety and press freedom
With the development of Artificial Intelligence (AI), online violence against journalists has taken a new turn. The online environment was already unsafe for journalists, but now AI acts as a catalyst for online harassment, intimidation, and subversion. This development not only hinders their work but also jeopardizes press freedom and the underlying democratic processes in general. Public awareness of this issue is needed to make progress in the areas of legislation and enforcement, as the role that AI plays in online violence has not yet been universally recognized by law at the European level. With this in mind, Free Press Unlimited delved into the world of artificial intelligence. In what ways is AI being used to attack journalists? What does AI-facilitated violence look like today? And what patterns can be identified?
AI facilitates existing forms of online violence against journalists, such as identity fraud, threats, and sexual violence. That is why we refer to “AI-facilitated violence” in this article. This includes so-called deepfakes: AI-generated videos, photos, or audio clips that are difficult to distinguish from real footage. Free Press Unlimited identified a total of 58 incidents within the MMF Monitor in which AI was used to obstruct or intimidate journalists. The MFRR’s Mapping Media Freedom (MMF) Monitor is a European database where journalists report incidents in which they were targeted. The incidents all took place in Europe between 2022 and 2026.
AI-Facilitated Scams
One primary form of AI use is the so-called AI-facilitated “scam.” In this scenario, the journalist’s reputation or expertise is exploited to promote a scam. Seventeen of the 58 incidents investigated fall into this category. This development undermines the credibility and expertise of journalists. In the fintech sector, where new technology is being developed for the financial world, deepfake scams rose by 700% in 2023.
One example is the case of Gabriel Mellqvist, a Swedish journalist who covers financial news. An AI-generated fake advertisement for an investment platform pretended that he was endorsing it. As a result, approximately 500 investors were defrauded, with a total loss of about 500 million Swedish kronor (approximately 46 million euros). The journalist himself says he is frustrated that his reputation has been damaged by this incident.
AI-Facilitated Russian Disinformation
A second category consists of Russian disinformation campaigns. These campaigns use personas of journalists (and sometimes entire news websites) to spread pro-Russian or anti-Ukrainian narratives. These disinformation campaigns are known by various names, such as Portal Kombat, Storm 15-16, Matryoshka (Overload), and Doppelganger. Sixteen of the 58 cases investigated can be linked to Russian disinformation strategies.
The case of French news anchor Julien Fanciulli, who works for France24, illustrates how one of these campaigns, Portal Kombat (also known as the “Pravda network”), operates. In 2024, a deepfake video of Fanciulli was published by pravda.fr and further disseminated by various pro-Russian information portals. In the deepfake video, Fanciulli, appearing as a France24 news anchor, reported that President Macron had postponed his trip to Kyiv due to a planned assassination attempt on the president. This narrative casts doubt on Western confidence in Ukraine by portraying Kyiv as unsafe.
The French intelligence agency VIGNIUM describes Portal Kombat as a network of news portals comprising at least 193 affiliated websites. They do not produce original content but re-post information from pro-Russian social media accounts and news channels to reach a Western audience. In addition, Storm 15-16, Matryoshka, and Doppelganger are characterized as “information laundering.” This means that disinformation is posted and disseminated on a massive scale, in part thanks to AI, so that its origin becomes unrecognizable.
Since July 2023, more than 8 million fake articles have been re-posted by the Pravda Network, according to the monitoring website portal-kombat.com, an initiative of the fact-checking organization Checkfirst. In addition, Matryoshka is particularly known for spreading AI deepfakes. Doppelganger mimics official news websites, sometimes using AI-generated content. Behind these websites are organizations funded by the Russian government.
Sexual Violence
A third category consists of sexual deepfakes. In these cases, AI is used to generate sexual images of journalists without their consent. Five of the 58 analyzed incidents fall into this category. Deepfakes are increasing rapidly. In 2023, there were an estimated 500,000 deepfakes online; according to a forecast, this number rose to 8 million by 2025. As many as 96–98% of all deepfakes consist of non-consensual sexual content online, with women being the victims in 99% of cases. The societal consequences of online gender-based violence are far-reaching. For example, previous research has shown that women may withdraw from the public sphere or be unable to continue practicing their profession.
Free Press Unlimited observes that the dissemination of this type of imagery can damage a journalist’s reputation or lead to further online, and sometimes physical, intimidation. The psychological harm suffered by victims can be comparable to that caused by physical sexual violence. Furthermore, AI-generated images are difficult to remove from the internet. These images can continue to cause harm long after their publication.
With relatively little effort, Free Press Unlimited found websites featuring explicit deepfakes of thirteen Dutch women, including three journalists. For privacy reasons, we have decided not to disclose the URLs of these websites. Under Dutch law, the distribution of non-consensual sexual deepfakes is prohibited. Nevertheless, illegal content featuring Dutch journalists is still online as of the writing of this article. In May 2026, NRC revealed that AI-generated porn websites are also hosted on Dutch territory, such as the website “Motherless.”
In Italy, too, two well-known female journalists and presenters fell victim to this. Francesca Barra and Selvaggia Lucarelli found AI-generated nude images of themselves on a website with millions of users that features multiple (well-known) women. Although the Italian police launched an investigation after this incident came to light, this website is also still online.
AI-facilitated political intimidation
The final category consists of journalists who have been targeted by an AI-facilitated political attack. They are known for their political convictions and have often been the target of physical or online violence in the past. The intimidation has a clear political motive. In addition, these journalists are more frequently the target of physical or online threats or violence. Five of the 58 incidents fall into this category.
The incident involving Serbian journalist Dinko Gruhonjić is a typical example, in which his position within a local press freedom organization is believed to be the cause. Since 2024, he has also been a key figure in the student protests in Novi Sad, Serbia. He was targeted by an AI-manipulated video in which it appeared as though he was saying he was happy to share a Croatian commander with a notorious war criminal. The video was also shared by legitimate media outlets.
Following the video, a series of new incidents involving Gruhonjić were reported to the MFRR, including death threats. It was not clear exactly who was behind the manipulated video in 2024. However, it is notable that nationalist actors and parties in particular are capitalizing on the video and accusing him of treason.
In some cases, AI-generated content is barely recognizable, if at all. That is why we at Free Press Unlimited make the following recommendations for better protection and legislation:
1. Be critical of online content. Who is creating it? Who stands to gain?
2. Use tools to identify and expose disinformation
3. Continue to report incidents to:
1. Monitoring organizations such as the MFRR
2. Relevant legal authorities (see an overview of the most important ones in Europe here)
3. Organizations that provide support for online abuse and violence, such as Offlimits.
Legal Framework
- Harmful deepfakes currently fall outside the scope of European legislation. Furthermore, at the EU level, the Digital Services Act only requires VLOPs (Very Large Online Platforms) such as X and Facebook to remove illegal content as quickly as possible. However, these incidents often occur outside of VLOPs, for example in Telegram channels. In addition, national legislation still determines what constitutes illegal content.
- In addition to EU legislation, steps have also been taken at the national level to address AI-facilitated violence. Denmark recently adopted an amendment to its national copyright law making unauthorized deepfakes a violation of every individual’s right to their likeness and voice. This blueprint has the potential to be adopted as legislation in other European countries as well.
- In the Netherlands, too, progress has been made toward criminalizing AI-facilitated violence. The lawsuit filed by TV host Welmoed Sijtsma against the creator of a non-consensual deepfake paves the way for legal prosecution of similar incidents in the future. The perpetrator was sentenced to 180 hours of community service. Under Dutch law, creating an explicit deepfake is therefore a criminal offense, provided that the footage appears so realistic that it is not immediately clear at first glance that it involves AI.
