Paragraph 1
Generative artificial intelligence is advancing at breakneck speed, and with it comes an unprecedented flood of synthetic content that is becoming increasingly difficult to distinguish from reality. Across the internet, AI-generated images, videos, audio clips, and text are proliferating, raising urgent questions about the integrity of public information and the potential for large-scale disinformation. While advocates hail the productivity gains and creative possibilities of these tools, critics warn that the same capacities enabling them are also eroding the shared foundation of factual reality. At the University of Georgia, scholars who study both propaganda and artificial intelligence are sounding alarms that the chaos seen today is only the beginning. Roger Stahl, a professor of communication studies and expert in propaganda and media systems, describes the current moment as the edge of a much larger transformation. “We’re kind of walking into this house of mirrors,” Stahl said, “and we’re just seeing the very, very edge of it.” His metaphor captures a disorienting landscape in which the line between authentic and fabricated content has blurred beyond easy recognition. Stahl and his colleagues argue that the societal consequences of this shift are not merely hypothetical; they are already visible in manipulated political narratives, fake news stories, and viral hoaxes that move faster than any verification system can catch. The problem, they contend, is not simply that AI can create convincing fakes, but that the modern information environment has become uniquely vulnerable to them. The combination of sophisticated generative tools and a fragmented, trust-depleted media ecosystem has produced a perfect storm for disinformation, one that will likely intensify before it improves.
Paragraph 2
To understand why AI-generated disinformation has become so potent, Stahl traces the roots back more than a decade before the latest wave of generative tools. He points to the rise of social media around 2010 as a turning point when the flow of information shifted away from centralized outlets such as national television networks and major newspapers. That transition did not merely change how people consumed news; it fundamentally altered the architecture of public trust. In the old media order, editorial boards and professional journalists served as gatekeepers, setting standards for what could be published and providing a measure of accountability. As audiences scattered across platforms, influencers, and unverified pages, those institutional safeguards began to lose authority. What emerged, Stahl argues, is a system in which credibility is no longer tied to editorial responsibility but to identity, social circles, and personal affinity. “There’s a kind of limbic, instinctual trust in whatever personalities people happen to be following at the time,” he said. That emotional, identity-based trust is powerful and difficult to correct. When a person receives information from someone they already admire or identify with, they are far less likely to scrutinize the information’s source or authenticity. AI accelerates this dynamic by filling the fragmented information ecosystem with synthetic content that looks credible enough to spread widely. It no longer takes a sophisticated campaign to manufacture plausible disinformation; almost anyone can generate it on demand. As authorship becomes less visible and less relevant, Stahl notes, the very concept of accountability recedes. “Authorship becomes a lot less clear, transparent and maybe less important,” he observed, meaning that even if a piece of content is exposed as false, tracing it to a responsible party can be nearly impossible.
Paragraph 3
Kyle Johnsen, a professor in UGA’s College of Engineering and associate director of the university’s Institute for Artificial Intelligence, approaches the problem from a technical perspective. He agrees that the threat is both real and underappreciated, especially by those who have not kept pace with rapid advances in the field. The same underlying technology behind chatbots and image generators has made producing disinformation almost effortless, Johnsen argues. Models can now generate lifelike human faces, clone voices, and compose text that reads as if written by a journalist or policy expert. The technical ceiling keeps rising faster than public understanding can follow. Johnsen cautions that knowledge about AI is perishable; what may have seemed impossible just a few months ago is now routine. “Your knowledge about AI, if it is more than three months out of date, you might think AI is not capable of something, but it actually is,” he said. In this environment, even sophisticated observers can underestimate what is real and what is fabricated because they are relying on outdated assumptions about machine capabilities. Detection tools exist, and they are improving, but Johnsen stresses that these tools only function when someone is actively looking for synthetic content. The overwhelming majority of internet users do not run every image through a detector or question every video frame. Most people, he says, can still tell the difference between an AI-generated photo and a real one, but only under circumstances that invite careful inspection. “Most people can still tell the difference between an AI-generated photo and a real one, but it’s not something you’re just going to accept,” Johnsen explained. “You’d have to be looking for it.” Passive consumption, in other words, is exactly what AI disinformation exploits.
Paragraph 4
Given the scale of the problem, policymakers have begun exploring regulatory responses, but the scholarly consensus is that these efforts are far from sufficient. One of the most prominent proposals is mandatory labeling of AI-generated content, which has gained traction especially in Europe. The European Union’s AI Act includes transparency requirements for synthetic media, obliging developers and deployers to disclose when content is AI-generated. In principle, labels would allow audiences to distinguish authentic material from fabricated material and prevent deceptive use of deepfakes or fake news. Stahl views labeling as a meaningful starting point, acknowledging that it sets a necessary baseline for public awareness. But Johnsen is more skeptical, arguing that labeling alone is unrealistic and unlikely to succeed without serious enforcement. He warns that bad actors have every incentive to ignore labeling requirements; a disinformation campaign designed to manipulate audiences is not going to voluntarily mark itself as synthetic. “We can’t just put policies out there and hope for the best, especially when we’re dealing with something that is really, really difficult,” Johnsen said. The problem lies in the structural realities of the internet: platforms are global, content can be created in one jurisdiction and shared in dozens of others, and anonymous actors can evade accountability with relative ease. Even if regulators require labels, there is no guarantee that the most dangerous content will carry them. Technical enforcement mechanisms, such as watermarking and encrypted provenance metadata, are being developed but are not yet universally adopted, and they can often be stripped or circumvented. The gap between policy aspiration and practical enforcement remains a major concern for both scholars.
Paragraph 5
For Stahl, the longer-term answer lies not in technical fixes or labeling mandates alone but in a revival of trustworthy, accountable journalism. He noted that established news organizations still possess something that synthetic content and anonymous bad actors do not: accountability. A credible outlet has a reputation to protect, a staff of professionals, and legal and ethical obligations to correct errors. Even in an era of declining trust, those institutions remain the best hope for sorting fact from fiction. Stahl lamented that the shift from legacy media to social media necessarily involved the delegitimization of the old system, creating a profoundly fractured audience that is insulated in its own enclaves. People no longer share a common reference point; they consume information tailored to their existing beliefs and social networks. In such a fragmented landscape, AI-generated disinformation can easily reinforce preexisting biases and further isolate audiences from authoritative sources. “There’s a profoundly fractured audience that are kind of insulated in their own little enclaves. And the move from a centralized media legacy system to social media system necessarily entails the delegitimization of the old system,” Stahl said. Yet he insists that the solution remains what it always has been: organizations with credibility, clout, and something to lose. “The answer has always been established media organizations,” he added. The challenge, then, is twofold. Media institutions must find ways to rebuild public trust and make themselves credible to audiences who were taught to doubt them. At the same time, the public must be educated to understand how newsrooms operate, why editorial standards matter, and how to evaluate the reliability of sources they encounter online.
Paragraph 6
Taken together, the perspectives from Stahl and Johnsen highlight the need for a multi-pronged response to AI-powered disinformation. There is no single solution that will undo the damage already done to the information ecosystem. Technical research on AI detection and content provenance will continue to matter, but such tools require widespread adoption and active use to be effective. Policy frameworks like the EU’s AI Act can establish norms and provide legal recourse, but enforcement mechanisms must be strengthened to prevent labels from becoming merely cosmetic. Meanwhile, rebuilding the authority of established journalism may be the most difficult task of all, because it involves restoring trust in a public that has been conditioned to be skeptical of institutional voices. Stahl and Johnsen agree that the arrival of generative AI has fundamentally changed the landscape, but they also see room for vigilance and resilience. The problem is not that AI is inherently destructive; it is that the social and cultural checks on misinformation have failed to evolve at the same speed as the technology. Universities and research centers like UGA’s Institute for Artificial Intelligence have a role to play in advancing public understanding and developing new safeguards. Journalists, too, must adapt their practices to account for the ease with which images and statements can be fabricated. Ultimately, the fight against AI disinformation is not merely a technical problem to be solved with better algorithms, or a legal problem to be addressed with new laws. It is a civic challenge requiring citizens to become more careful consumers of information and requiring institutions to earn back the trust they have lost. The house of mirrors Stahl describes is still growing, but the path through it depends on human judgment, accountability, and a renewed commitment to truth.
