The Fragmentation of Truth

Guest Blogger

Valérie Bélair-Gagnon

When we talk about AI and fact-checking, we often fixate on the informational: the deepfake, the viral lie, or the bot. Yet the disinformation crisis is fundamentally institutional. We have reached a crossroads where we must shift our focus from the viral lie to the underlying political economy that shapes who defines truth, and at what cost. If we fix the information but leave the infrastructure of truth-making in the hands of a few market-driven empires, we have not solved the disinformation crisis; we have simply automated it. 

This institutional struggle is at the heart of the anti-disinformation assemblage, a contingent, often messy configuration of platforms, states, technology organizations, and editorial actors. In an ongoing collaborative book project, my co-authors and I argue that this assemblage is currently undergoing a profound fragmentation. These diverse actors are held together by a struggle for definitional authority: the power to decide what constitutes a social problem and what requires an intervention.

The Response-Side Gap

If you look at the last decade of interdisciplinary research on disinformation, there is a massive supply-side bias. We have thousands of papers on how disinformation is produced and why people believe it, but there is a lacuna on the response side. We need to know more about how the institutions responding to the crisis actually allocate authority in practice. 

Governance interventions in this field often reinforce existing hierarchies. When we introduce AI into this mix, we are adding a tool for efficiency and automating existing institutional biases. This occurs in the automation of harm detection over veracity. During high-stakes elections, platforms deploy AI to suppress borderline content, material that violates policy, and risks brand reputation. This shifts the goal from a shared pursuit of truth to a mechanical pursuit of market stability. AI tools are programmed to find what is least disruptive to a platform's advertising ecosystem. We are thus moving from state-led propaganda to platform-led digital governance, in which the authority to verify information has shifted from public bodies to private, algorithmic entities. 

Three Pillars of the AI-Truth Economy

As we explore this shift, three critical questions emerge:

1. Does AI reshuffle or entrench power? Most AI tools are beholden to the walled gardens of platforms. If a fact-checking startup builds an AI detection tool, its survival depends on API access granted by Meta or Google. In this context, AI centralizes verification infrastructure rather than democratizing it.
2. Is the goal truth or market stability? AI moderation is implemented because it is scalable and cost-effective, not because it is the most accurate. We are seeing an epistemological relativism where “harm,” which carries legal and brand risks, is prioritized over “veracity.”
3. Has “truth” become a luxury good? We are currently in what our team identifies as the "Retraction Era" (2022–2025). Platforms are scaling back human trust and safety teams in favor of AI to reduce costs. By failing to mandate "human-in-the-loop" oversight, law and policy have allowed a global decoupling: while the Global North retains some algorithmic protections, the Global South is left with automated-only moderation. 

Consider the Tigray War in Ethiopia. While English-language content enjoys layers of human and algorithmic oversight, internal documents, such as the Facebook Papers, revealed that Meta’s AI systems were blind to languages like Amharic and Oromo. Inflammatory calls for ethnic violence remained active for days because AI tools lacked the linguistic nuance to identify the threat. Platforms prioritized the high-cost maintenance of "truth" in Western markets while leaving the Global South to be moderated by black-box systems that could not recognize the significance of the language until violence had already spilled into the streets. 

Scaling the Analysis: Macro, Meso, and Micro

To understand how this functions, we must look at the anti-disinformation assemblage at three levels:

*       Macro Level: Three digital empires are currently projecting power. The U.S. follows a neoliberal model, privileging free markets; the EU acts as a regulatory superpower focusing on rights; and China utilizes a state-driven model of surveillance. AI is the technical and legal force currently shaping the boundaries of acceptable speech globally.

*       Meso Level: Initiatives such as Vera.ai and Logically Intelligence have framed disinformation as a technical problem, solvable with software. This technologizing of fact-checking turns an epistemic struggle into a data-management task. Similarly, X’s Community Notes shifts the labor of truth-seeking onto unpaid users, turning a political debate into a ranking problem.

*       Micro Level: Consider the human fact-checker. Unlike traditional journalists, fact-checkers make explicit epistemic judgments. When President Biden claimed his uncle was eaten by cannibals, legacy outlets reported the claim without evidence but hesitated to label it a lie; independent fact-checkers like Snopes made an explicit judgment. Yet, as these actors partner with platforms, they become serfs, a precarious labor force for the very tech giants they are meant to monitor. 

Moving Past the Disinformation Crisis

We must recognize that AI and law are not just tools; they are reflections of a crisis of trust in professional authority. 

To counter the walled gardens of the digital empires, we must move toward a public-interest infrastructure. Regulators should mandate that platforms provide real-time API access to independent researchers. To stop global decoupling, we must regulate the quality of AI, not just its output quantity, perhaps by mandating specific ratios of human local experts. 

Law and policy can regulate bad content while simultaneously addressing the asymmetries of truth-making. If we do not address the infrastructure, we have not solved the crisis. We have automated it.

 Valérie Bélair-Gagnon is Associate Professor at the Hubbard School of Journalism and Mass Communication, University of Minnesota-Twin Cities. You can reach her by e-mail at vbg@umn.edu.

Collaborators on the forthcoming book project include: Steen Steensen (OsloMet), Rebekah Larsen (MIT), Lucas Graves (Universidad Carlos III de Madrid), Bente Kalsnes (Kristiana University College), Oscar Westlund (OsloMet/Gothenburg), Lasha Kavtaradze (Kristiana University College), and Reidun Samuelsen (Norwegian Media Authority).