Fewer expressions prove that words don’t mean anything anymore like “post-truth“. The sentence “We live in the post-truth era” has been repeated enough times, in multiple languages and on endless news outlets that we are now desensitised to it and its implications. We should have known, back in January 2017, when Kellyanne Conway, then counsellor to U.S. president Donald J. Trump, gave that infamous interview to NBC Meet the Press, defending then White House Press Secretary Sean Spicer’s false claims about the crowd size at Trump’s inauguration. Upon being asked by moderator Chuck Todd why the president had sent out his press secretary to utter a provable falsehood “in front of the public”, she first dodged the question and then, when pressed, replied “you’re saying it’s a falsehood, and they gave… our Press Secretary Sean Spicer gave alternative facts to that”. The very fact that Chuck Todd didn’t immediately call Conway out on the blatant absurdity of that sentence and that we, as a society, accepted that “alternative facts” were now an option should have warned us about the slippery slope that would have brought us, in recent years, to battle what we call FIMI and DIMI (respectively, “Foreign Information Manipulation and Interference” and “Domestic Information Manipulation and Interference“).
Around that same time, as if there weren’t enough signs that acquiring neologisms from the Trump administration was a terrible idea, the president started using the term “fake news” to designate any piece of information that didn’t fit into his narrative and this expression, which was weak to begin with, lost any credibility and any power in pointing out what once would have been called, more concisely and effectively, lies.
Those of us who deal or concern ourselves with the accuracy of information, the ethics of content production, and with analysing and understanding the way in which humans share and develop knowledge now have to contend with something way more pernicious than mere “lies” or even “alternative facts” or “fake news”. Because the gap between FIMI and disinformation (or misinformation) is about as vast as the difference in severity between a twisted ankle and stage four bone cancer. Where the former makes you hobble your way through a few press conferences, the latter fatally compromises the very structure of your body, leading to your inevitable demise.
And while the expression “Fake news” stuck around for longer than it had any right to do, security agencies, researchers, and policy institutions have long been aware of how imprecise it is, not to mention the fact that it was weaponised by the very actors it was meant to describe and eventually hollowed out through overuse. The replacement terminology is both more technical and more honest, but a lot less catchy: the concepts of Foreign Information Manipulation and Interference (FIMI) and Domestic Information Manipulation and Interference (DIMI) herald in a fundamentally different understanding of what information warfare is, who wages it, and why the old mental models, the ones in which so many of us have participated enthusiastically, featuring fact-checkers and media literacy campaigns, are structurally insufficient responses to what is now a security threat on an unprecedented scale.
In 2025, the European External Action Service (EEAS, the Diplomatic Service of the European Union) detected 540 FIMI incidents globally, involving 10,500 social media channels and websites producing or amplifying manipulative content, with attacks increasing in frequency, intensity, and coordination. Over a hundred countries were targeted and so were over a hundred high-profile individuals, including heads of state, and close to 200 organisations ranging from NATO to traditional media outlets and academic institutions. This is no longer a matter of misinformed individuals sharing dubious links on Facebook and has nothing to do with your boomer uncle sending you sensational fake-news in the form of memes in the family Whatsapp group. This is the deadly phase of that same disease and it looks much more akin to actual warfare, albeit conducted through narrative rather than conventional weapons.
From “fake news” to FIMI
The “fake news” framing carries an implicit and dangerously misleading assumption: that the problem be content. Correct the false claim, remove the offending post, label the misleading headline, and the threat recedes. We know, by now, that this is absolutely never the case. This phenomenon has been studied extensively both within the humanities and by neuroscience and debunking, while useful to keep the record straight for those who care, has consistently failed to cause significant change to the opinions and courses of action of those whose Weltanschauung has been shaped by deliberately false, misleading, and mostly polarised information.
The concept of FIMI was developed, by contrast, as an attempt so systematise a complex phenomenon that can’t be reduced to the concept of “lies” or “propaganda”. FIMI is defined not by what is said but by how and why it is disseminated and to what extent. Whereas disinformation and misinformation describe false or misleading content that anyone can produce, willingly or unwillingly, FIMI describes coordinated manipulative behaviour by foreign actors (or, in the case of DIMI, domestic ones) using deceptive methods to interfere in a specific society’s information environment, specifically affecting the flow of information that said society is exposed to. In other words, this is not like dumping a few barrels of ink into a stream, this is like going to back to the spring the river originates from, digging down to the water reserve it taps into, taking out all the water and replacing it with ink and sewage. In this scenario, adding a purifier downstream would be a futile and slightly ridiculous effort, even if there were still some water to be distilled from the constant stream of slop.
This distinction matters enormously. A piece of content can be technically accurate and still function as FIMI if it is selectively amplified by coordinated inauthentic networks to exploit an existing social fracture. Conversely, outright fabrications may be irrelevant to FIMI analysis if they lack the coordinated infrastructure behind them.
Information manipulation goes well beyond false content and includes amplification of genuine but partial information, amplification of genuine information in contexts where it is likely to be polarising and where accurate analysis is unlikely to happen, and of course censorship, which is precisely why governments, researchers, and security agencies are increasingly relying on the concept of FIMI. That is, if and when they are interested in opposing it, of course. The focus shifts from policing narratives and fact-checking content to analysing behaviour: networks of accounts, infrastructure linkages, patterns of coordinated amplification, and the strategic objectives these patterns serve.
That makes FIMI analysis closer, methodologically, to cybersecurity and intelligence than to journalism or fact-checking. The tools required are network mapping, behavioural analysis, quantitative analysis and attribution frameworks, not editorial judgement or the painstaking tracking of facts and figures. In other words, we can no longer fight lies simply by telling the truth (which doesn’t mean we, as individuals, should stop engaging actively with the truth. More on that later).
The architecture of modern information manipulation: AI, scale, and the Doppelgänger problem
Perhaps nothing illustrates the evolution of FIMI more starkly than the absorption of artificial intelligence into its operational infrastructure. And this is not meant to disparage AI tout court. The fact is that, whether we agree with it or not, AI is a power-tool, an accelerator and a formidable instrument for scaling up anything. A content producer can scale up their production tenfold, a hundred-fold by using AI. A lier (intended in this context as any subject, individual or institutional, that intentionally engages in lying), will be able to lie faster, more often, for longer, and with wider exposure, if not altogether better. Within the context of the 4th EEAS Report on FIMI Threats, those described above as “liars” are called, rather less romantically and more effectively “threat actors”. The same report tells us that AI-generated text, synthetic audio, and manipulated video have shifted from experimental use to routine deployment, becoming cost-effective and scalable tools for said threat actors.
In 2025, one in four detected incidents by the EEAS involved the use of AI tools to produce or distribute content. That figure represents a sharp increase from the previous year, and the trajectory points relentlessly upward. While you should absolutely read the report, which will be linked at the end as one of the main sources for this document, it is also worth taking a short step back to express one caveat. The EEAS focused on the threats and threat actors that more strongly target Europe and they come, for obvious reasons, from outside of the EU. The nations that are most often found the be the points of origin of these threats aiming at influencing the geopolitical situation in Europe are Russia and China. This is a numeric fact, backed by data that can be analysed and verified by the reader. This does not mean, however, nor is it my understanding, that other national or multinational powers do not engage in FIMI or DIMI, nor does it mean that European Countries do not experience instances of FIMI and DIMI that are not connected to Russia or China. For sheer scale, it would be frankly surprising if two such superpowers were not in a leading position in any phenomenon connected to the geopolitical balance of the world. In this context, however, we will not be dissecting the responsibility of any individual nation, pointing fingers, or drawing rankings of friends and enemies. This article focuses exclusively on the phenomenon itself, on the fact that it exists and on the way it works. The point being that the very existence of FIMI is a threat to humanity, regardless of who uses it against whom.
Back to FIMI and AI. LLMs dramatically lower the cost-per-narrative and remove the bottleneck of human content production, meaning bots have finally taken the job of professional trolls and producers of what we used to call “fake news”. One could argue that you can’t even get paid to make up lies anymore: a robot stole your job. There is both poetic justice and the early sketch for a decent film plot there.
The so-called Doppelgänger campaigns offer a particularly unsettling illustration of how this works in practice. These campaigns blend covert tactics such as emotional manipulation, AI-generated content, cross-platform coordination, and information suppression to create content that mimics legitimate media sources. Fake websites cloning the visual identity of respected newspapers, AI-generated articles indistinguishable in appearance from genuine reporting, fabricated social media personas with years of synthetic posting history. The aggregate effect is not to persuade audiences of any particular falsehood so much as to make the information environment feel untrustworthy and hard to navigate. It is worth saying that this particular technique, while impressive, tends to generate less engagement than other FIMI IMS (Information Manipulation Set, defined by EEAS as “a collection of adversarial behaviours, tools, TTPs, and resources that is presumed to originate from the same threat actor”).
There is also a more insidious vector that has to do specifically with the way LLMs work: the deliberate corruption of AI training data in order to make sure that the most popular and widely used AI platforms are primed to do the threat actors’ work for them. Large Language Models are being “groomed” by operators who flood the internet with false information in an effort to manipulate results when people use these tools as if they were search engines (which they are not). The implication is profound: actors engaged in FIMI don’t really need to compete in the present information space, because they are attempting to shape the epistemological infrastructure through which future generations will access and synthesise knowledge and, so far, they have been successful in their attempts. This specific behaviour is slow-acting and almost entirely invisible until the contamination becomes apparent in AI outputs. In other words, while individuals throughout society are stripped of their competence and their ability to acquire, parse, and analyse information independently, and are enticed, seduced, or cajoled into being increasingly dependent on LLMs telling them what things are and how they work, suggesting patterns of actions and thought, while they are enthusiastically encouraged to drink from one stream and one stream only, someone is up there, pumping ink and sewage into the aquifer.
And it would be outstandingly naive to think of any individual government or institution being held accountable for poisoning our collective well. Of the 10.500 channels involved in FIMI incidents in 2025, according to the report, the vast majority (90,5%) consisted of covert assets linked to or aligned with a threat actor, with 86% classified as “state-aligned” rather than directly state-controlled. This architecture of deniability is essential to understanding why attribution remains so difficult and why the problem resists straightforward regulatory solutions.
Cultural awareness and DIMI
One aspect of FIMI that is worth considering is the fact that it can’t be “cookie-cut”. Better said, it does get scaled, but the output does lose quality as a result of being massively AI-produced and it would be a mistake for any threat actor to package content once and send it out indiscriminately to multiple foreign targets. In order to be effective, FIMI requires an in-depth knowledge of the social, demographic, cultural, and political outline of its target, especially when it is used to manipulate the public political discourse before an election.
Inevitably, this makes DIMI, Domestic Information Manipulation and Interference potentially more effective, since threat actors, in this case, are supposedly part of the same human environment as their targets. Not that this makes the preliminary research any less important, but it definitely changes the balance of powers and it also implies a different perspective, on the part of the threat actor, in terms of goals, if not of choice of content.
In several countries DIMI and FIMI are at the very least interconnected or even part of the same offensive, in that interference by domestic actors who manipulate information and act against democratic principles often involves a considerable amount of foreign influence, being channelled through domestic actors to amplify their actions and statements.
Generally speaking, foreign actors manipulating an election must build or recruit local distribution networks. A domestic actor already has authentic social capital, local credibility, access to real political communities. The Romanian elections in 2024 and the 2025 German federal election, according to the report, both illustrated how AI-powered domestic disinformation campaigns, using bots, fake media, and AI-generated content to fuel societal divides, can operate alongside or in loose coordination with foreign infrastructure without being reducible to it.
DIMI tactics include creating fake websites, producing AI-generated deepfakes, fabricating news, discrediting sources, and mobilising bot networks, with objectives ranging from swaying elections and gaining political traction to damaging the reputation of opponents and eroding trust in national institutions. Hungary’s weaponisation of state media (at least during the Orbán era), Slovakia’s legal constraints on NGOs, and Romania’s documented AI-based campaign interference are among the clearest recent examples. What makes these cases particularly troubling is that much of this activity, short of outright fraud, is not illegal. DIMI exploits the very openness of democratic systems against them.
No, you can’t fight FIMI (on your own)
As much as it costs to accept this, conscious prosumer behaviour and media literacy are not remotely enough to deal with this kind of threat. Such tools were once relevant and useful and they still hold value on an individual level, but they are simply no longer effective on a societal level. Advising individuals to verify sources, check for emotional manipulation in headlines, consult fact-checking websites, and cultivate media literacy is not wrong, of course, it is just inadequate, if compared to the sheer magnitude of what FIMI can achieve.
Consider the scale: in 2025, approximately 43,000 pieces of content, i.e. texts, audio, and video, were recorded across 19 different platforms, with 88% of instances concentrated on the platform X alone. Can you fact check 43,000 lies? Can anyone? If any of them were thrown at you right now, if any of them, after being packaged in a respectable and credible way, landed on your national news media, would you be able to recognise it for what it is? Would you even know to check for inconsistencies? And, even if you did, how many, out of millions, would even consider it? It is easy to see why these campaigns are successful. Once the burden of proof is on you, the audience, and the production of content can be scaled to un precedented volumes, what can human-driven fact-checking achieve? The sheer volume of content generated by AI-assisted FIMI operations overwhelms any individual capacity for verification. An attentive, media-literate citizen, diligently cross-referencing every suspicious claim against multiple authoritative sources, is not competing with a handful of foreign trolls working night shifts in a basement (and even that was an unfair fight, if the trolls were getting paid for full working days and the user was fact-checking in their own downtime). They are competing with automated systems capable of producing content in dozens of languages simultaneously, adapting narratives in real time to breaking events.
Nearly half of recorded FIMI incidents in 2025 were triggered by specific events, particularly breaking-news moments and electoral processes, which provide fertile ground for information manipulation. High-emotion, high-stakes moments, when individuals are most motivated to seek information and most cognitively stretched, appear to be the moments when FIMI is most intensively deployed. Asking citizens to be more sceptical during a crisis is, arguably, asking them to perform a task that is neurologically at its most difficult precisely when it is most necessary.
If a solution can be found at all to this momentous problem, it surely has to be structural rather than individual. Most major social media platforms restrict access to data that would allow assessment of the magnitude of information manipulation activities. Without that data, civil society organisations, researchers, and even national security agencies are working with partial information. This also makes the case for a different distribution of responsibility. However much we might find it infuriating when friends, acquaintances, and strangers on the internet take what to us is obvious bait and espouse conspiracy theories or spread incorrect information, we have to acknowledge that individual citizens, navigating this environment without institutional support, are mostly not failing through inattention or insufficient education, but rather operating inside a system whose architecture was not designed with their epistemic protection in mind. In other words: no, it’s not your fault, no, you are not stupid or lazy (well, you might be, I don’t know you, but, if you are, this specific failure is not proof or your lack of mental prowess), this is simply beyond you, because you are not a bot, you are a human who has to focus on other things too.
Is there anything we can do, as a society, to counter FIMI? Maybe.
The emerging institutional response
The 2026 report marks a meaningful shift in institutional thinking. While previous editions established the analytical methodology, the response framework, and the exposure matrix, this document focuses, finally, on deterrence. It even introduces a FIMI Deterrence Playbook as a structured approach aimed at increasing the costs and reducing the operational space for those conducting information manipulation.
The Playbook’s logic is borrowed from established security doctrine: rather than attempting to counter every individual incident reactively, identify and attack the infrastructure that makes the operations possible. Essentially, try and take away their ink barrels before they can dump them in your stream. Or at least make them uncomfortable, heavy, hard to manoeuvre, make the ink water-soluble, switch it for strawberry flavouring, do anything, in short, to make life as hard for threat actors as they are doing for society. Identifying key vulnerabilities within the infrastructure, intermediaries, and supply chains used by threat actors is an excellent starting point, because it allows domestic operators (in the case of this particular report, that would be the EU and its partners) able to mobilise existing instruments such as sanctions, law enforcement cooperation, digital regulation, and resilience-building to disrupt these activities.
This is the “kill chain” approach applied to information warfare and it can look like different solutions, of varying degrees of efficacy. For instance, a nation could target the domain registrars hosting fabricated news sites, making them inaccessible from the affected country (even though users would still be able to access them using VPNs, with the added bonus of being able to brag about being silenced and repressed), sanction the financial intermediaries funding covert amplification networks, even engage law enforcement where FIMI activities intersect with actual cyber crime, or regulate platforms under instruments like the Digital Services Act to enforce transparency around coordinated inauthentic behaviour. The board has actually come up with the first publicly available interactive dashboard on information manipulation activities, called the FIMI Explorer, which maps networks and key players, serving as a resource for the wider FIMI defender community.
Electoral interference remains the clearest immediate priority. Russia targeted elections in Germany, Poland, Romania, Moldova, the Czech Republic, and Côte d’Ivoire in 2025, and the same playbook deployed during Moldova’s parliamentary elections has been redeployed in Armenia ahead of its June 2026 elections. The pattern is predictable enough that forward-looking defence is possible and the report makes it clear that this anticipatory posture is now central to the EU’s approach.
For DIMI, the challenge is harder. Domestic manipulation does not always cross the threshold of illegality, and democratic societies are, correctly, wary of giving governments broad powers to police political speech, as that would be extremely dangerous to any democracy aspiring to stay one. A domestic approach, therefore, calls for more structural responses, ranging from stronger independent media to enforced platform transparency, from adequately funded and provenly independent public service broadcasting to and legal frameworks that distinguish between protected political expression and coordinated inauthentic manipulation. None of these are quick fixes and no country can currently boast a full, exhaustive, and operative set of measures to this end.
The report itself says it best: “A stronger deterrence posture is […] required not to escalate, but to ensure accountability. Those who attempt to manipulate the information environment should face constraints, exposure and an increase in costs. Making such activities more difficult, less effective and riskier for perpetrators is central to protecting democratic systems.”
Still quoting directly from the report, the playbook focuses on deterrence, as a move to anticipate the threat and make it harmless before it develops. Specifically, it suggests the following:
● Raise costs and risks for threat actors, ensuring that manipulation no longer represents a favourable cost–benefit calculation.
● Isolate and constrain operational space, narrowing the room for manoeuvre.
● Target critical capabilities, striking at the “nerves” of the FIMI architecture, including organisational networks, financial flows, operators and technical infrastructure.
● Push perpetrators to reshape their strategy, compelling adaptation, increasing uncertainty and forcing them to waste time and resources trying to rebuild infrastructure.
● Delegitimise threat actors, reducing their credibility and influence.
Personally, I am surprised the last point was not implemented earlier, as it appears to me as a perfect example of the kind of game that can work both ways, but I guess that might have something to do with the epistemological crisis we have been experiencing for years, as a carefully planned result of the massive wave of anti-intellectualism that the success of populist party politics has brought about. There is currently a widespread tendency to distrust any source connected with institutional and peer-reviewed knowledge, which is deemed to be “impure” and enslaved to the powers that be, while blind trust is encouraged towards anonymous and non-verifiable sources, as long as they speak the same anti-intellectual and power-disparaging language that their audiences have learned to trust.
Having said all that, the report leaves some room for the kind of advocacy I have been pursuing over the years. Specifically, it dedicates a paragraph to the importance of building resilience, in order to make FIMI attacks less effective within our society. That also means improving our individual and collective ability to “recognise, assess and respond to manipulation”. This should be done, according to EEAS, through a combination of “public exposure, strategic communication, media literacy and capacity-building across institutions and civil society”.
Which is, I guess, our clue to keep doing what we have been doing, and advocate for media literacy, user awareness of manipulation techniques, and more transparent media practices. We just have to be aware that we are no longer the main actors on the frontline and we probably never were.