We Are Fighting a Hydra with a Sword. We Are Losing.
FARIDA KHALAF13 min read·Just now--
This article is a collaboration between Farida Khalaf and . It grew out of our conversation following a closed EU security seminar Mila attended. Her notes, her case analysis, and her documentation of four European operations are the evidentiary backbone of everything that follows. The architecture is ours. The evidence is hers.
The Immortal Machine: When Influence Operations Learn to Regenerate
In 1744, the Swiss naturalist Abraham Trembley cut a Hydra into pieces expecting it to die.
Instead, each piece grew into a completely new organism.
Trembley had not destroyed the creature. He had multiplied it. The Hydra, a translucent freshwater polyp barely visible to the naked eye, has no brain, no central nervous system, no headquarters. Its intelligence is distributed across its entire body. Every cell contains the instructions for rebuilding the whole. Cut it anywhere and the biology simply reorganizes around the damage, regenerating not just the wound but the complete organism from whatever remains.
Scientists have studied Hydra for three centuries. In laboratory conditions, some specimens have survived over 10,000 days without deteriorating. They show no signs of aging. They do not decline. They simply continue, renewing, reorganizing, rebuilding indefinitely.
Security researchers studying influence operations in the European Union have a name for the digital system that shares these properties.
They call it Cyborg.
What Cyborg Is Not
Before describing what Cyborg is, it is worth being precise about what it is not.
It is not a bot farm. The previous generation of influence operations ran on automated accounts coordinated by command-and-control servers, thousands of non-human actors flooding platforms with engineered content. That model became increasingly visible as platforms improved their detection of coordinated inauthentic behavior. Automated signatures could be identified. Servers could be seized. Accounts could be suspended at scale.
Cyborg replaced the machine with the human.
Not because humans are cheaper, in some configurations they are not. But because a human being posting a comment on Reddit is indistinguishable from a genuine user posting a comment on Reddit. A human-written Medium article passes every automated authenticity check because it is, in the narrow technical sense, authentic. A real person wrote it. The inauthenticity is not in the execution. It is in the instruction, and the instruction exists several layers removed from the content, in a closed chat that the executor never sees.
This is the architectural shift that makes Cyborg a different category of system from its predecessors. It did not improve the bot farm. It made the bot farm obsolete.
The Operating Environment
Cyborg’s theater of operations is the European Union and the choice is precise.
The EU presents a specific combination of conditions that makes it the ideal environment for a modular influence system: multiple languages, multiple regulatory cultures, multiple national media ecosystems, and a political structure in which a “concern” raised credibly in one member state can cascade institutionally into regulatory attention across the continent.
Bot farms were blunt instruments. They could flood a platform but could not navigate the difference between what is persuasive to a German audience and what is persuasive to a Romanian one. Cyborg was designed for exactly the environments where bluntness fails, where cultural calibration is the competitive advantage, and where the EU’s regulatory architecture can be used as an amplification mechanism rather than a constraint.
The primary channels are Facebook, Reddit, Quora, and Medium. Each serves a distinct function in the information cascade. Facebook generates the initial emotional detonation. Reddit provides community validation, the appearance of organic discussion among people who discovered the issue independently. Quora captures the search query from someone who heard about it and wants to understand, and becomes the indexed authority anchor. Medium provides the long-form breakdown that lends the narrative the appearance of investigative legitimacy.
Together, the cascade creates a specific perceptual effect: that everyone is talking about this, that it emerged independently across multiple platforms, that the pattern of concern is organic.
None of it is organic. All of it was coordinated from a closed chat.
The Architecture
Cyborg’s structure follows a biological logic that maps precisely onto its operational requirements.
The Brain: Distributed Intelligence
The intellectual center of Cyborg is not a content team. It is a multidisciplinary unit psychologists, economists, sociologists, specialists in behavioral modeling, whose function is to understand the target audience before a single piece of content is deployed.
What are the audience’s fears? What are its conformist pressures? What does it want confirmed? What counts as credible in this specific linguistic and regulatory environment?
The Brain’s output is a library: central theses, supporting narratives, a glossary of frames, a map of what works and what fails by platform and cultural context, and, critically, a set of safe formulations. Statements engineered to create maximum moral pressure with minimum explicit defamation. Technically defensible. Functionally destructive.
Like the Hydra’s distributed nervous system, the Brain contains no single point of failure. Its knowledge is distributed across specialists who can be replaced without losing the accumulated library.
The Nerves: Synchronization
Between the Brain and the executors sit two to three intermediaries, the only communication interface with the broader system. They relay commands, recruit handlers, filter information leaks, and dispatch tasks to executors who never see the full operational picture.
Above them sits the synchronizer: the orchestrator who sets the rhythm of operations, distributes roles across platforms, monitors execution, and adjusts course when the narrative encounters resistance.
The synchronizer is both the most valuable and the most vulnerable component of the system. Without it, the modules produce noise, independently generated content that lacks the coordinated rhythm that creates the illusion of organic consensus. The synchronizer is what turns a collection of disconnected posts into a coherent narrative that appears to have emerged from genuine public concern.
The Organs: Five Functional Modules
Intelligence continuously scans the information environment around the target, pain points, trust triggers, fear triggers, weak points in the target’s relationships with partners, investors, regulators, and media.
The Narrative Engine produces operational content: central theses, supporting cases, a glossary of frames, and the library of safe formulations. It does not produce defamation. It produces insinuation, moral pressure applied through implication, juxtaposition, and selective framing designed to raise questions without making verifiable claims.
Low-Context Executors are cheap freelancers who perform individual tasks without access to the full operational picture. They post, comment, compile, and format, each performing one node of a chain they cannot see end to end. Their value is scale, speed, and the diffusion of responsibility that makes attribution nearly impossible. Each individual action is defensible. The coordinated pattern is not.
Amplifiers transfer the central thesis across platforms in sequence, creating the perception of independent organic discovery. The cascade Facebook to Reddit to Quora to Medium is engineered to feel like a conversation that started spontaneously and spread naturally.
Wikipedia Operators work within the platform’s neutrality requirements to consolidate frames through encyclopedic space. They are not SEO practitioners. They are knowledge engineers making desired sources appear normal for a topic, using thematic pages as external anchors that lend the narrative the appearance of established fact. Wikipedia formally requires neutrality and reliable sources. Cyborg operates in the zone where the wording looks neutral and the sources look suitable, while the cumulative effect is directional.
Three Types of Operations
Reputational operations target trust. In attack mode: plant red flags that surface in due diligence searches, fix the target’s name to specific concerns in indexed content, create a toxic background that makes partners and investors instinctively cautious, without any single piece of content being demonstrably false. In defense mode: stabilize trust under attack, neutralize accusations, maintain a shield narrative that occupies the space a damaging story would otherwise fill.
Marketing operations target decisions. In positive mode: create conditions for deals and market entry through coordinated social proof that appears independent. In negative mode: generate doubt at the top of the competitor’s funnel, increase friction at the point of decision, undermine trust without making verifiably false claims.
Regulatory operations are the most architecturally sophisticated. The goal is institutional transfer: converting a reputational narrative into a regulatory one. Complaints filed with consumer protection bodies. Concerns raised through industry associations. Public signals of consumer harm amplified until they reach regulators who have a legal obligation to investigate. A complaint from one person is noise. A pattern of complaints from multiple independent sources across multiple member states is a regulatory event, even if those complaints were coordinated from a single closed chat.
Three Tempos
Operations run at three distinct speeds.
Flash: 24 to 48 hours. Capture the first explanation of an event before the target can establish its own narrative. Emotional framing. Speed over precision.
Burner: 2 to 4 weeks. Maintain pressure. Build plausibility through repetition and variation. Change angles of attack without changing the core thesis, so each new piece of content appears to be independent confirmation rather than coordinated amplification.
Background: up to 3 months. Fix the label in long-term formats, Q&As, evergreen breakdowns, indexed discussions, that outlive the news cycle and continue to surface in search results long after the active operation has ended.
A target facing a full Cyborg operation might experience it this way: doubt appears on Facebook over a weekend. Reddit discussions validate and extend it over two weeks. Quora answers cement it as the default search result for anyone researching the target. A Medium breakdown provides the appearance of investigative legitimacy. Wikipedia quietly reflects the new framing. Six months later, a due diligence search surfaces the narrative as established context and no single piece of content is obviously false.
The Evidence: Four European Cases
The following section is Mila Agius’s direct analysis of four documented European influence operations from 2024 to 2026. Her cases move the architecture described above from theory to evidence, each module named, each mechanism observed in operation.
The Illusion of Organic Consensus: How the Modules of the “Cyborg” System Use Infrastructure to Strip Away Political Agency
We have grown used to assessing information threats through the outdated lens of bot farms, where servers coordinate thousands of automated accounts. That frame is false. An analysis of European influence operations from 2024 to 2026 reveals a shift toward a decentralized infrastructure that, for the purposes of this analysis, can be described as the “Cyborg” system.
This is not a bot farm. It is a distributed architecture that combines machine and human actors, then turns supposedly neutral algorithms into free amplifiers. Four recent European cases show how the functional modules of this system embed themselves in local contexts, transforming a digital illusion into a real institutional crisis.
Module 1: Narrative Engine and Mimicry (Case: Doppelgänger in Poland and the EU)
Earlier generations of systems relied on blunt reach. “Cyborg” relies on cultural calibration. In the case of the Doppelgänger network, we can see the work of a Narrative Engine: the system did not produce generic spam, but localized anti-European and anti-Ukrainian messaging around the specific internal anxieties of Polish society.
Mechanism: Legitimacy is manufactured through brand impersonation. The network created fake media sites and lookalike domains, generating content that resembled authentic journalism. When attempts were made to block these resources, the system displayed a Hydra-like quality: through rapid domain replacement and resilient infrastructure, the network absorbed targeted hits and kept operating. Blocking one node does not destroy the system. It merely forces it to reorganize like a living organism.
Module 2: Environmental Saturation and AI Hijacking (Case: Portal Kombat / Pravda)
The most technically sophisticated stage of stripping away agency is the abandonment of direct persuasion in favor of altering the knowledge base itself. The Portal Kombat network scaled production to as many as 10,000 pieces of content a day across dozens of languages.
Mechanism: This is a classic capture of trusted environments. The point of mass publication was not for humans to read the material, but to seize search indexing and enter the body of content later picked up by AI systems, a process known as LLM grooming. As a result, links to Pravda resources began appearing in chatbot responses, being used as sources and serving to anchor information by creating an appearance of credibility. This exploits the layer of autonomous search crawlers and indexing systems, which collect and further circulate such content simply because it is algorithmically optimized to be found.
Module 3: Isolated Executors and Camouflage (Case: Moldova)
The perfect digital forgery is a real person. In Moldova, the coordinators of the campaign against Maia Sandu and European integration relied not only on automation, but also on the recruitment of local users through closed Telegram channels.
Mechanism: This is the operational function of Low-Context Executors. Real people are paid to perform narrow tasks such as posting a comment or sharing a text, without seeing the full architecture of the operation. To platforms and regulators, these actions look like organic public discontent. The system uses living people as a behavioral shield, delegating responsibility to them and substantially complicating any classical attribution of the attack.
Module 4: Institutional Transfer (Case: Romania)
The ultimate aim of any sophisticated influence operation is to force a digital metric to alter physical reality. In Romania, the algorithmic promotion of the previously little-known candidate Călin Georgescu through TikTok led to an extraordinary outcome: the Constitutional Court annulled the results of the presidential election.
Mechanism: A rupture between formally authentic interest and hidden coordination. The candidate’s digital campaign, while presenting itself as organic viral growth, was accompanied by major investigations into opaque funding, covert coordination, and the role of algorithmic amplification. This is the highest stage of the system at work: online dynamics are successfully converted into an institutional shock for the state.
Architecture of Vulnerability (The Verdict)
An analysis of these four operations reveals the central “decision handoff” moment. European regulators are trying to fight “Cyborg” with the metrics of a previous era by counting removed accounts and blocked servers.
But a decentralized architecture has no single server that can be switched off. Block the network on Facebook, and it will continue living on Telegram. Remove the executors, and the system will recruit new ones through closed chats. More than that, the act of blocking becomes content in itself: the system uses censorship as a signal for regeneration and as fuel for the next wave of outrage. As long as defense is built around the search for a single command center, institutions will keep losing to a system designed to survive without a head.
The Regeneration Problem
Mila Agius’s verdict names the mechanism precisely. What follows is where that mechanism leads.
When security teams identify a Cyborg operation and begin removing content, something specific happens: the operation does not stop. It reorganizes.
Suspend the Facebook accounts, the Reddit module continues. Remove the Reddit threads, the Quora answers remain indexed. Challenge the Wikipedia edits, new editors arrive with cleaner sourcing and more carefully neutral wording. Identify the synchronizer, new intermediaries are recruited from closed chats within days. Prosecute the freelancers, they never knew the full operation anyway, and replacements cost less than the legal proceedings.
More precisely: the disruption itself becomes content. “Why are they trying to silence this conversation?” The censorship narrative feeds the next wave. The intervention generates the next Flash tempo.
Like the Hydra, Cyborg does not merely survive damage. It uses damage as a regeneration signal. The attack is the stimulus. The removal is the instruction to rebuild.
This is the property that makes Cyborg architecturally different from its predecessors. A bot farm, when disrupted, is disrupted. The server goes down. The operation ends. Cyborg, when disrupted, grows back — because it has no center to remove, no server to seize, no single point whose elimination stops the whole.
The Loop Function
In cybersecurity, certain classes of virus contain what is called a loop function, a self-rewriting script that overrides detection rules, mutates its own signature, and rewrites itself before the antivirus can recognize what it is looking at.
By the time the defense identifies version one, version two is already running. By the time it catches version two, version three has overwritten the rules that made version two detectable. The virus is not evading detection. It is rewriting the definition of what detection is looking for.
Cyborg’s 2026 evolution incorporates this principle at the narrative level.
The safe formulations, the insinuations engineered to create maximum moral pressure with minimum explicit defamation are now being continuously recalibrated by AI systems against platform moderation rules. The moment a formulation pattern is flagged, the loop rewrites it. Not a human rewriting it. An automated system doing it continuously, faster than any moderation team can respond, against moderation rules that themselves change slower than the rewriting loop runs.
The narrative mutates. The signature changes. The operation continues.
The Crawlbot Layer
The final escalation layer is the one that requires no instruction at all.
AI agents, crawlbots designed to scan information environments for content matching certain parameters discover Cyborg-planted content organically. They were not instructed to find it specifically. They found it because it was engineered to be found: the right keywords, the right engagement patterns, the right platform signals that tell an automated amplification system that this content is worth spreading.
The crawlbot reposts, summarizes, cross-references, and embeds the narrative in other content streams. Each amplification appears independent because it is. The crawlbot has no knowledge of Cyborg. It is not a participant in the operation. It is a neutral system doing what it was built to do: surface and spread content that matches its parameters.
At this point the operation has achieved something that has no precise precedent in the history of influence operations: a human-seeded narrative being autonomously amplified by machines that have no awareness they are weapons, at a speed and scale no human coordination could match, across platforms and search indexes simultaneously.
The human swarm planted the seed. The machine harvested it. The Hydra regenerated through silicon.
And it runs continuously. Twenty-four hours a day. Seven days a week. Rewriting itself against detection rules faster than the rules can be updated. Amplifying through crawlbots that cannot be held responsible because they have no intent. Regenerating around every intervention because the intervention is the signal to rebuild.
What This System Actually Is
Cyborg as documented in four European cases and as it is evolving in 2026 represents the convergence of three distinct threat architectures into one operational system.
The Hydra model: distributed, decentralized, immortal through regeneration. No center to remove. No head to cut.
The virus loop function: self-rewriting, self-mutating, overriding detection rules faster than they can be updated. No signature to catch.
The crawlbot amplification layer: autonomous spread through neutral systems with no knowledge of the operation they are serving. No instruction to trace.
Three systems. One operation. Zero attribution.
The question this architecture raises is not whether it can be stopped through conventional means. It cannot, for the same reason the Hydra cannot be killed by cutting it. Every intervention produces more of what the intervention was trying to eliminate.
The question is what a detection and response architecture looks like when the threat has no server, no signature, no center, and no off switch and when the systems amplifying it have no awareness they are doing so.
That question does not yet have a complete answer.
Disclaimer: The four European cases documented in Mila Agius’s section — Doppelgänger, Portal Kombat, Moldova, and Romania are drawn from her direct seminar notes and analysis. The Hydra framework, loop function, and crawlbot dynamics represent the authors’ analytical synthesis of the seminar material and documented 2025–2026 AI and influence operation trajectories.
Originally published at https://fafi25.substack.com.
