This reaction is understandable.
Disinformation presents itself as content, so we respond to it as content. A post, a video, a meme, a headline. Something visible, shareable, measurable.
The problem is that this framing no longer matches the reality of how influence operations actually function.
What we increasingly face is not a series of misleading messages, but a system designed to shape the environment in which decisions are made. The focus has shifted from persuading people of specific claims to interfering with how societies make sense of events at all.
When influence operates at that level, responding one post at a time becomes not just insufficient, but structurally irrelevant.
This shift has begun to surface in institutional thinking.
NATO now explicitly frames modern influence as a form of cognitive warfare: activity aimed at disrupting human decision-making below the threshold of armed conflict, affecting civilians and institutions alike.
The target is not belief in a single proposition, but the cognitive processes through which individuals and societies observe, interpret, and act. In simple terms, it is about interfering with sense-making.
That distinction matters. Democracies do not collapse because citizens believe one falsehood. They weaken when the shared processes that allow disagreement, verification, and judgment begin to erode.
When people no longer trust what they see, no longer know which signals matter, and no longer feel capable of orienting themselves in complex situations, political agency decays. Confusion becomes the outcome.
This helps explain why content moderation alone has struggled to deliver meaningful results.
Platforms remove material, yet manipulation persists. Governments impose sanctions, yet operations continue. Investigations expose networks, yet new ones quickly appear. The problem is not a lack of awareness.
It is a mismatch between the level at which the threat operates and the level at which responses are applied.
To understand that mismatch, it helps to stop looking at disinformation as speech and start examining it as infrastructure.
Recent work by EU DisinfoLab has provided one of the clearest frameworks for doing so.
Instead of treating influence operations as isolated campaigns or attributing every incident directly to a state actor, the report introduces the concept of the Information Manipulation Set, or IMS.
An IMS is not a single operation or organization. It is a recurring configuration of behaviors, tools, infrastructure, narratives, and intermediaries that can be reused across multiple campaigns.
This distinction is crucial.
At the tactical level, researchers observe individual posts, videos, or fabricated websites.
At the operational level, those incidents are linked through shared hosting providers, amplification networks, content formats, or narrative patterns.
At the strategic level, these operational clusters can sometimes be attributed to sponsoring actors and long-term objectives.
The mistake, the report argues, is collapsing all three levels into one.
When every viral incident is treated as its own campaign, the result is duplication, noise, and inflated threat perception without strategic clarity.
Multiple investigations end up describing the same underlying system under different names, overwhelming policymakers while obscuring where real leverage exists.
Seen through the IMS lens, disinformation begins to resemble a supply chain rather than a conversation. Content is produced, hosted, amplified, monetized, and recycled through a web of intermediaries.
Financial sponsors, technical service providers, cloaking services, advertising networks, and platform features all play roles. Remove one post, and the system continues functioning. Disrupt the supply chain, and the operation becomes costly or unsustainable.
This perspective also exposes an uncomfortable reality: many of the weakest points in influence operations lie not abroad, but within democratic systems themselves.
The DisinfoLab report documents how sanctioned actors have continued to access hosting, distribution, and platform services inside the European Union, despite formal restrictions.
In theory, asset freezes prohibit such access. In practice, enforcement has been inconsistent, fragmented across national authorities, and easily circumvented through proxies.
The gap, as the report notes bluntly, is not intelligence. It is execution.
This matters because influence operations thrive on asymmetry.
They are relatively cheap to run, highly tolerant of failure, and designed to operate at scale.
Defensive responses, by contrast, are slow, legally cautious, and politically constrained. Each new requirement for attribution, consensus, or proportionality increases the cost of disruption, while the cost of launching another operation remains low.
Until recently, that asymmetry still had one important limitation: human labor. Influence operations required people to write posts, manage accounts, coordinate timing, and adapt narratives. Even large operations faced constraints of speed, endurance, and attention.
That constraint is now disappearing.
A 2026 paper published in Science describes the emergence of malicious AI swarms as a new phase in information warfare.
These systems combine large language models with autonomous agents capable of maintaining persistent identities, coordinating behavior, adapting to engagement in real time, and operating continuously across platforms.
The danger is not that AI can generate false text. That capability already exists. The deeper shift is operational. AI removes friction.
It enables influence systems that do not tire, do not require salaries, and do not rely on centralized command structures that are easy to detect. Thousands of personas can be adjusted dynamically, testing variations, refining narratives, and amplifying what works at machine speed.
Where traditional bot networks relied on repetition and synchronization, AI-driven systems can mimic human heterogeneity. They vary tone, timing, and style. They learn from feedback. They infiltrate communities gradually rather than flooding them. Influence becomes persistent rather than episodic.
In this environment, manipulation no longer looks like an attack. It looks like background noise — a steady shaping of norms, language, and perceived consensus.
Over time, this can produce what the researchers describe as synthetic consensus: the illusion that certain views are widely held, not because people genuinely agree, but because coordinated systems manufacture social signals at scale.
This matters because democratic reasoning depends heavily on perceived peer norms. People do not evaluate every claim independently. They infer credibility from what others appear to accept. When those signals are engineered, independence erodes.
The result is not mass persuasion, but gradual distortion of the informational environment itself.
What makes this especially dangerous is the convergence between existing influence supply chains and emerging AI capabilities.
The infrastructure mapped through IMS frameworks does not disappear when automation increases. It becomes more valuable. Hosting providers, platform features, recommendation systems, and intermediaries become force multipliers rather than bottlenecks.
In other words, tomorrow’s influence operations will not replace today’s systems. They will accelerate them.
This creates a serious problem for governance.
Regulatory approaches built around identifying actors, attributing campaigns, and moderating content struggle in dynamic environments where behavior adapts continuously. Attribution becomes slower just as operations become faster. Enforcement remains national while operations remain transnational.
By the time one system is documented, it has already evolved.
It is tempting, at this point, to place responsibility entirely on social media platforms.
Their design choices, business models, and incentive structures undeniably shape the information environment. Engagement-driven ranking rewards emotionally charged content. Synthetic activity can inflate metrics that translate into revenue. Users cannot easily distinguish authentic participation from coordinated behavior.
Yet focusing solely on platforms risks repeating the same mistake at a different level.
Influence operations exploit existing systems; they do not create them. Political actors, domestic and foreign alike, benefit from ambiguity. Market incentives align poorly with epistemic health. And democratic societies are rightly cautious about empowering authorities to police truth.
The problem, then, is not simply technological. It is systemic.
What emerges from these three bodies of research — NATO’s cognitive framing, the IMS operational model, and the analysis of AI-enabled manipulation — is a shared conclusion not often stated directly: defending democracy at the level of content is no longer sufficient.
Effective responses must focus on systems rather than messages.
That means visibility into infrastructure, accountability for intermediaries, consistent enforcement of existing rules, and transparency around coordinated behavior rather than semantic judgment of speech. It means prioritizing provenance and patterns over individual claims.
It also means investing in civic capability.
Not persuasion, but recognition. Citizens do not need to be told what to think. They need tools and literacy that allow them to recognize when the environment itself is being shaped.
Understanding that influence is organized, persistent, and structural restores a measure of agency. Confusion thrives in isolation; clarity emerges from pattern recognition.
None of this offers a simple fix.
Influence operations will not disappear. Automation will continue. The information environment will remain contested.
The question is not whether manipulation can be eliminated, but whether its costs can be raised and its effects contained.
At its core, democracy depends on shared sense-making. Disagreement is not a flaw; it is the mechanism.
But when the processes that allow people to observe, orient, decide, and act are systematically disrupted, political choice becomes hollow.
We are no longer facing a problem of false content. We are facing the emergence of cognitive infrastructure designed to interfere with how societies think.
The choice, then, is not between regulation and free speech, or technology and democracy. It is between continuing to fight symptoms — posts, memes, incidents — or finally addressing the systems that produce them.
Only one of those approaches matches the reality we now inhabit.
And now, you know.
Sources
1. NATO Science & Technology Organization
Chief Scientist Research Report: Cognitive Warfare
NATO STO, 2025.
This report outlines NATO’s current scientific understanding of cognitive warfare, defining it as activity aimed at influencing and disrupting human decision-making below the threshold of armed conflict. It examines the cognitive dimension of modern conflict, links influence operations to sense-making and decision processes, and emphasizes the need for whole-of-society resilience.
2. EU DisinfoLab
Building a Common Operational Picture of Foreign Information Manipulation and Interference (FIMI)
EU DisinfoLab, 2025.
This report introduces the Information Manipulation Set (IMS) framework, providing an operational model for understanding how influence operations are structured, sustained, and reused across multiple incidents. It focuses on infrastructure, intermediaries, and enforcement gaps rather than isolated content or attribution alone.
3. Schroeder et al.
How malicious AI swarms can threaten democracy
Science, 2026.
This peer-reviewed paper analyzes how autonomous AI agents can enable persistent, adaptive, and large-scale influence operations. It explains how AI-driven coordination removes traditional human constraints on influence campaigns, increases the difficulty of detection and attribution, and amplifies existing vulnerabilities within democratic information ecosystems.
