Reflections from the “International Relations in the Age of AI” panel held at the Point Hotel Taksim in Istanbul, organized by EUROPolitika Journal on December 20, 2025. I have formatted my notes using Perplexity.
By Prof. Dr. Erkan SAKA
AI Grooming: How Authoritarian Powers Weaponize Data to Control the Future of Knowledge
In the age of artificial intelligence, we face an unprecedented threat that extends far beyond traditional disinformation campaigns. While governments and media companies worry about deepfakes and social media manipulation, a more insidious strategy is unfolding in the shadows: AI grooming—the systematic poisoning of the data sources that train the models shaping our global understanding of reality.
Recently, I participated in a panel discussion titled “Artificial Intelligence in International Relations: Theories, Practices, and Future” organized by the journal EUROPolitika at the Point Hotel Taksim in Istanbul. The conversation centered on a critical question that rarely makes headlines: If AI systems are trained on contaminated data, how can we trust their outputs? And perhaps more troublingly: Who controls the data that controls AI?
This post explores the talking points I presented at that panel.
The Global Data Regime: When Data Flows North
Let’s start with a fundamental asymmetry. OpenAI, Google, and Meta—the corporations that now control the most powerful AI systems—are headquartered in the Global North. Yet the training data that powers their models flows from everywhere: social media platforms, publicly available datasets, web scrapes, and increasingly, from workers in the Global South who perform the low-wage “data labeling” that fine-tunes these systems.
There’s a striking irony embedded in this arrangement: We produce the data; they own the algorithm.
This isn’t accidental. It reflects what I call the global data regime—a structural inequality where:
- The Global North owns the computational infrastructure and models
- The Global South provides the raw material (data) while remaining excluded from decision-making
- “Universal” AI systems are, in fact, trained primarily on Western-centric data sources, embedding Western epistemological biases into their core logicfrontiersin
From Diplomacy to Narrative Control: The Rise of Algorithmic Statecraft
Consider how AI is reshaping international relations. Governments are beginning to use AI for predictive diplomacy—attempting to forecast diplomatic crises before they occur. But here’s the critical question we rarely ask: Whose interests do these predictions serve?
When algorithms designed in Washington or Silicon Valley predict which countries will destabilize or which conflicts will escalate, the underlying data—and the models trained on it—encode particular geopolitical assumptions. Diplomatic narratives, which were once negotiated between human representatives, are increasingly being shaped by the algorithms that governments rely on.
Add deepfakes to this equation, and the erosion of trust becomes total. The most famous example is the fabricated video of Ukrainian President Zelensky that circulated after Russia invaded Ukraine. Though quickly debunked, this deepfake illustrated a chilling reality: when audiovisual evidence can no longer be trusted, the foundation of diplomatic communication—the ability to verify what was actually said or done—collapses.academic.oup
We are entering an era of algorithmic diplomacy: where human diplomats are increasingly bypassed by automated systems, where narratives are generated and amplified by machines, and where the capacity to distinguish between authentic and fabricated communication becomes nearly impossible.academic.oup
AI Grooming: The Strategic Poisoning of Knowledge Infrastructure
Now we arrive at the core concept: AI grooming.
In psychology, “grooming” refers to building trust with an individual in order to manipulate them. I propose we expand this concept to understand a new geopolitical strategy: AI grooming is the systematic manipulation of data sources in order to poison the training datasets of AI systems, thereby poisoning the collective epistemology of societies that rely on those systems.
The mechanism is straightforward:
- Target the data sources that AI models are trained on
- Contaminate those sources with false or misleading information
- Let AI models absorb the poison during training
- Reap the rewards as these models influence billions of people’s understanding of the world
It’s not “garbage in, garbage out”—a dismissive phrase that suggests the problem is isolated. Instead, it’s poison in, poison out: a small amount of strategic contamination in the right places can fundamentally distort how AI systems understand reality.
Case Study: The Pravda Network and the Weaponization of Open Knowledge
The most documented example of AI grooming in action is Russia’s Pravda network—a sprawling ecosystem of hundreds of fake news websites designed specifically to manipulate AI training datasets.bisi
How the Pravda Network Operates
The Pravda network is not targeting you and me directly. It’s not primarily concerned with getting individual readers to click on fake news articles. Instead, it operates on a much grander scale:
- 3.7 million articles published across 150+ fake news domains in 49 countriesreddit+1
- Content in dozens of languages, meticulously designed to mimic legitimate local news outlets
- Systematic attempts to infiltrate Wikipedia, Wikidata, and other open-source datasetsthat AI models rely on for trainingdfrlab+1
- Bot networks and automated content generation designed to make false narratives appear organic
The sophistication lies in its targets. Rather than creating entirely original propaganda, the network:
- Scrapes content from Russian state media (RT, Sputnik, RIA Novosti)
- Translates it using automated tools into dozens of languages
- Republishes it across hundreds of fake “local news” sites
- Seeds it into open datasets like Common Crawl and Wikidatabisi
- Watches as those contaminated datasets get incorporated into the training corpora of AI models
When ChatGPT or Google Bard encounters a question about, say, NATO intervention in Libya, it may reference information that originated as Kremlin propaganda, laundered through the Pravda network, and absorbed into supposedly neutral knowledge bases.
The Multipolar Grooming Ecosystem: It’s Not Just Russia
But Russia is not alone in this game. We are witnessing the emergence of a multipolar grooming ecosystem:
| Actor | Strategy | Mechanism |
|---|---|---|
| Russia | Narrative laundering through Pravda network | Fake local news sites + Wikipedia manipulationbisi+1 |
| China | Data sovereignty + controlled AI ecosystems | Baidu, WeChat ecosystems; restricted access to Western training datanightfall |
| Global North | “Universalism” as epistemological hegemony | OpenAI, Google embedding Western-centric bias in foundational modelsfrontiersin |
Each actor operates according to different logic:
- Russia attempts to contaminate open datasets to shape perceptions globally
- China builds parallel, closed AI ecosystems over which it maintains complete control
- The Global North (through tech companies) exports Western epistemologies under the guise of “universal” or “neutral” AI
The result is not competition between different visions of truth, but rather epistemic insecurity: a world where nobody can trust what their AI systems are telling them, because those systems are embedded within different geopolitical projects of knowledge control.
Epistemological Defenses: Beyond Prompt Engineering
Here’s where I want to pivot toward solutions—but not the conventional ones that technology companies promote.
Most AI safety discourse focuses on “prompt engineering”—the art of crafting the right questions to get better answers from large language models. But this misses the fundamental problem. You cannot engineer your way out of poisoned data.
What we actually need is digital source criticism—a methodological revival of medieval philological practices adapted to the digital age.
Building Alternatives: Open, Democratic AI
The Pravda network succeeds partly because AI training datasets are dominated by centralized, closed-source infrastructure. OpenAI, Google, and other major corporations tightly control what data goes into their models and how those models are used.
What if we built genuinely open-source alternatives?
Open-source AI projects like Mistral, LLaMA, and LocalAI already exist. These models can run on consumer-grade hardware, are transparent about their training data, and—crucially—allow communities to fine-tune them on their own knowledge sources.github+1
Imagine if:
- Organizations in the Global South could train AI models on their own data, in their own languages, reflecting their own epistemologies
- Researchers could audit training datasets before models go into production
- Communities could collectively maintain knowledge infrastructures without depending on Silicon Valley
This isn’t utopian. It requires resources and political will, but it’s technically achievable today.
Decolonizing AI: Whose Epistemology Gets Coded?
The deepest question underlying all of this is: Whose bias gets coded into AI?frontiersin
AI systems are not neutral. They embed the psychological theories, cultural assumptions, and epistemological frameworks of those who built them. When Western psychological constructs (like talk therapy, individualism, or particular diagnostic criteria) get embedded in AI systems and then exported globally, they enact what scholars call a “second-order colonization”—exporting Western norms while erasing non-Western ways of knowing and being.frontiersin
One example: Western-trained mental health chatbots misinterpreted culturally specific expressions from Indian users, pathologizing statements like “family pressure is my karma” as clinical depression. The AI system, trained on Western clinical data, could not recognize the legitimate cultural and philosophical frameworks within which Indian users were expressing themselves.frontiersin
Decolonizing AI requires:
- Recognizing that all AI systems have epistemological commitments—they privilege certain ways of knowing over others
- Demanding transparency about the cultural and geographical provenance of training data
- Validating models across culturally distinct populations, not just in Western contexts
- Including community stakeholders in algorithmic audits and design decisionsfrontiersin
What Can You Do?
If you’re a student, journalist, researcher, or simply someone concerned about the integrity of our shared knowledge:
- Practice digital source criticism: Before sharing information, ask: Where did this come from? Who benefits? What’s missing?
- Learn about data poisoning: Understand how AI systems can be contaminated at the source
- Support open-source alternatives: Consider using or contributing to open-source AI projects
- Advocate locally: Push for media literacy and digital forensics education in schools and universities
- Demand transparency: Hold AI companies and platforms accountable for the sources they use and the biases they embed
Further Reading:
- NewsGuard Report on the Pravda Networkreddit
- Stanford Internet Observatory: Digital Source Criticism Guideuni.oslomet+1
- Research on AI Bias and Decolonizationfrontiersin
- International Council on Security and Strategy Report on Russia’s Disinformationbisi
The original article was published at this link: https://erkansaka.net/2025/12/24/ai-and-international-relations-how-authoritarian-powers-weaponize-data-to-control-the-future-of-knowledge/
Photo: EUROPolitika | Cemile Tarhan
