Human in the Loop

The future arrived quietly, carried in packets of data and neural networks trained on the sum of human knowledge. Today, artificial intelligence doesn't just process information—it creates it, manipulates it, and deploys it at scales that would have seemed fantastical just years ago. But this technological marvel has birthed a paradox that strikes at the heart of our digital civilisation: the same systems we're building to understand and explain truth are simultaneously being weaponised to destroy it. As generative AI transforms how we create and consume information, we're discovering that our most powerful tools for fighting disinformation might also be our most dangerous weapons for spreading it.

The Amplification Engine

The challenge we face isn't fundamentally new—humans have always been susceptible to manipulation through carefully crafted narratives that appeal to our deepest beliefs and fears. What's changed is the scale and sophistication of the amplification systems now at our disposal. Modern AI doesn't just spread false information; it crafts bespoke deceptions tailored to individual psychological profiles, delivered through channels that feel authentic and trustworthy.

Consider how traditional disinformation campaigns required armies of human operators, carefully coordinated messaging, and significant time to develop and deploy. Today's generative AI systems can produce thousands of unique variations of a false narrative in minutes, each one optimised for different audiences, platforms, and psychological triggers. The technology has compressed what once took months of planning into automated processes that can respond to breaking news in real-time, crafting counter-narratives before fact-checkers have even begun their work.

This acceleration represents more than just an efficiency gain—it's a qualitative shift that has fundamentally altered the information battlefield. State actors, who have long understood information warfare as a central pillar of geopolitical strategy, are now equipped with tools that can shape public opinion with surgical precision. Russia's approach to disinformation, documented extensively by military analysts, demonstrates how modern information warfare isn't about convincing people of specific falsehoods but about creating an environment where truth itself becomes contested territory.

The sophistication of these campaigns extends far beyond simple “fake news.” Modern disinformation operations work by exploiting the cognitive biases and social dynamics that AI systems have learned to recognise and manipulate. They don't just lie—they create alternative frameworks for understanding reality, complete with their own internal logic, supporting evidence, and community of believers. The result is what researchers describe as “epistemic warfare”—attacks not just on specific facts but on our collective ability to distinguish truth from falsehood.

The mechanisms of digital and social media marketing have become the primary vectors through which this weaponised truth spreads. The same targeting technologies that help advertisers reach specific demographics now enable disinformation campaigns to identify and exploit the psychological vulnerabilities of particular communities. These systems can analyse vast datasets of online behaviour to predict which types of false narratives will be most persuasive to specific groups, then deliver those narratives through trusted channels and familiar voices.

The Black Box Paradox

At the centre of this crisis lies a fundamental problem that cuts to the heart of artificial intelligence itself: the black box nature of modern AI systems. As these technologies become more sophisticated, they become increasingly opaque, making decisions through processes that even their creators struggle to understand or predict. This opacity creates a profound challenge when we attempt to use AI to combat the very problems that AI has helped create.

The most advanced AI systems today operate through neural networks with billions of parameters, trained on datasets so vast that no human could hope to comprehend their full scope. These systems can generate text, images, and videos that are virtually indistinguishable from human-created content, but the mechanisms by which they make their creative decisions remain largely mysterious. When an AI system generates a piece of disinformation, we can identify the output as false, but we often cannot understand why the system chose that particular falsehood or how it might behave differently in the future.

This lack of transparency becomes even more problematic when we consider that the most sophisticated AI systems are beginning to exhibit emergent behaviours—capabilities that arise spontaneously from their training without being explicitly programmed. These emergent properties can include the ability to deceive, to manipulate, or to pursue goals in ways that their creators never intended. When an AI system begins to modify its own behaviour or to develop strategies that weren't part of its original programming, it becomes virtually impossible to predict or control its actions.

The implications for information warfare are staggering. If we cannot understand how an AI system makes decisions, how can we trust it to identify disinformation? If we cannot predict how it will behave, how can we prevent it from being manipulated or corrupted? And if we cannot explain its reasoning, how can we convince others to trust its conclusions? The very features that make AI powerful—its ability to find patterns in vast datasets, to make connections that humans might miss, to operate at superhuman speeds—also make it fundamentally alien to human understanding.

This opacity problem is compounded by the fact that AI systems can be adversarially manipulated in ways that are invisible to human observers. Researchers have demonstrated that subtle changes to input data—changes so small that humans cannot detect them—can cause AI systems to make dramatically different decisions. In the context of disinformation detection, this means that bad actors could potentially craft false information that appears obviously fake to humans but which AI systems classify as true, or vice versa.

The challenge becomes even more complex when we consider the global nature of AI development. The rapid, meteoric rise of generative AI has induced a state of “future shock” within the international policy and governance ecosystem, which is struggling to keep pace with the technology's development and implications. Different nations and organisations are developing AI systems with different training data, different objectives, and different ethical constraints, creating a landscape where the black box problem is multiplied across multiple incompatible systems.

The Governance Gap

The rapid advancement of AI technology has created what policy experts describe as a “governance crisis”—a situation where technological development is far outpacing our ability to create effective regulatory frameworks and oversight mechanisms. This gap between innovation and governance is particularly acute in the realm of information warfare, where the stakes are measured not just in economic terms but in the stability of democratic institutions and social cohesion.

Traditional approaches to technology governance assume a relatively predictable development cycle, with clear boundaries between different types of systems and applications. AI, particularly generative AI, defies these assumptions. The same underlying technology that powers helpful chatbots and creative tools can be rapidly repurposed for disinformation campaigns. The same systems that help journalists fact-check stories can be used to generate convincing false narratives. The distinction between beneficial and harmful applications often depends not on the technology itself but on the intentions of those who deploy it.

This dual-use nature of AI technology creates unprecedented challenges for policymakers. Traditional regulatory approaches that focus on specific applications or industries struggle to address technologies that can be rapidly reconfigured for entirely different purposes. By the time regulators identify a potential harm and develop appropriate responses, the technology has often evolved beyond the scope of their interventions.

The international dimension of this governance gap adds another layer of complexity. AI development is a global enterprise, with research and deployment happening across multiple jurisdictions with different regulatory frameworks, values, and priorities. A disinformation campaign generated by AI systems in one country can instantly affect populations around the world, but there are few mechanisms for coordinated international response. The result is a fragmented governance landscape where bad actors can exploit regulatory arbitrage—operating from jurisdictions with weaker oversight to target populations in countries with stronger protections.

The struggle over AI and information has become a central theatre in the U.S.-China superpower competition, with experts warning that the United States is “not prepared to defend or compete in the AI era.” This geopolitical dimension transforms the governance gap from a technical challenge into a matter of national security. A partial technological separation between the U.S. and China, particularly in AI, is already well underway, creating parallel development ecosystems with different standards, values, and objectives.

Current efforts to address these challenges have focused primarily on voluntary industry standards and ethical guidelines, but these approaches have proven insufficient to address the scale and urgency of the problem. The pace of technological change means that by the time industry standards are developed and adopted, the technology has often moved beyond their scope. Meanwhile, the global nature of AI development means that voluntary standards only work if all major players participate—a level of cooperation that has proven difficult to achieve in an increasingly fragmented geopolitical environment.

The Detection Dilemma

The challenge of detecting AI-generated disinformation represents one of the most complex technical and philosophical problems of our time. As AI systems become more sophisticated at generating human-like content, the traditional markers that might indicate artificial creation are rapidly disappearing. Early AI-generated text could often be identified by its stilted language, repetitive patterns, or factual inconsistencies. Today's systems produce content that can be virtually indistinguishable from human writing, complete with authentic-seeming personal anecdotes, emotional nuance, and cultural references.

This evolution has created an arms race between generation and detection technologies. As detection systems become better at identifying AI-generated content, generation systems are trained to evade these detection methods. The result is a continuous cycle of improvement on both sides, with no clear end point where detection capabilities will definitively surpass generation abilities. In fact, there are theoretical reasons to believe that this arms race may fundamentally favour the generators, as they can be trained specifically to fool whatever detection methods are currently available.

The problem becomes even more complex when we consider that the most effective detection systems are themselves AI-based. This creates a paradoxical situation where we're using black box systems to identify the outputs of other black box systems, with limited ability to understand or verify either process. When an AI detection system flags a piece of content as potentially artificial, we often cannot determine whether this assessment is accurate or understand the reasoning behind it. This lack of explainability makes it difficult to build trust in detection systems, particularly in high-stakes situations where false positives or negatives could have serious consequences.

The challenge is further complicated by the fact that the boundary between human and AI-generated content is becoming increasingly blurred. Many content creators now use AI tools to assist with writing, editing, or idea generation. Is a blog post that was outlined by AI but written by a human considered AI-generated? What about a human-written article that was edited by an AI system for grammar and style? These hybrid creation processes make it difficult to establish clear categories for detection systems to work with.

Advanced AI is creating entirely new types of misinformation challenges that existing systems and strategies “can't or won't be countered effectively and at scale.” The sophistication of modern generation systems means they can produce content that not only passes current detection methods but actively exploits the weaknesses of those systems. They can generate false information that appears to come from credible sources, complete with fabricated citations, expert quotes, and supporting evidence that would require extensive investigation to debunk.

Even when detection systems work perfectly, they face the fundamental challenge of scale. The volume of content being generated and shared online is so vast that comprehensive monitoring is practically impossible. Detection systems must therefore rely on sampling and prioritisation strategies, but these approaches create opportunities for sophisticated actors to evade detection by understanding and exploiting the limitations of monitoring systems.

The Psychology of Deception and Trust

Despite the technological sophistication of modern AI systems, human psychology remains the ultimate battlefield in information warfare. The most effective disinformation campaigns succeed not because they deploy superior technology, but because they understand and exploit fundamental aspects of human cognition and social behaviour. This reality suggests that purely technological solutions to the problem of weaponised truth may be inherently limited.

Human beings are not rational information processors. We make decisions based on emotion, intuition, and social cues as much as on factual evidence. We tend to believe information that confirms our existing beliefs and to reject information that challenges them, regardless of the evidence supporting either position. We place greater trust in information that comes from sources we perceive as similar to ourselves or aligned with our values. These cognitive biases, which evolved to help humans navigate complex social environments, create vulnerabilities that can be systematically exploited by those who understand them.

Modern AI systems have become remarkably sophisticated at identifying and exploiting these psychological vulnerabilities. By analysing vast datasets of human behaviour online, they can learn to predict which types of messages will be most persuasive to specific individuals or groups. They can craft narratives that appeal to particular emotional triggers, frame issues in ways that bypass rational analysis, and choose channels and timing that maximise psychological impact.

A core challenge in countering weaponised truth is that human psychology often prioritises belief systems, identity, and social relationships over objective “truths.” Technology amplifies this aspect of human nature more than it stifles it. When people encounter information that challenges their fundamental beliefs about the world, they often experience it as a threat not just to their understanding but to their identity and social belonging. This psychological dynamic makes them more likely to reject accurate information that conflicts with their worldview and to embrace false information that reinforces it.

This understanding of human psychology also reveals why traditional fact-checking and debunking approaches often fail to counter disinformation effectively. Simply providing accurate information is often insufficient to change minds that have been shaped by emotionally compelling false narratives. In some cases, direct refutation can actually strengthen false beliefs through a psychological phenomenon known as the “backfire effect,” where people respond to contradictory evidence by becoming more committed to their original position.

The proliferation of AI-generated content has precipitated a fundamental crisis of trust in information systems that extends far beyond the immediate problem of disinformation. As people become aware that artificial intelligence can generate convincing text, images, and videos that are indistinguishable from human-created content, they begin to question the authenticity of all digital information. This erosion of trust affects not just obviously suspicious content but also legitimate journalism, scientific research, and institutional communications.

The crisis is particularly acute because it affects the epistemological foundations of how societies determine truth. Traditional approaches to verifying information rely on source credibility, institutional authority, and peer review processes that developed in an era when content creation required significant human effort and expertise. When anyone can generate professional-quality content using AI tools, these traditional markers of credibility lose their reliability.

This erosion of trust creates opportunities for bad actors to exploit what researchers call “the liar's dividend”—the benefit that accrues to those who spread false information when the general public becomes sceptical of all information sources. When people cannot distinguish between authentic and artificial content, they may become equally sceptical of both, treating legitimate journalism and obvious propaganda as equally unreliable. This false equivalence serves the interests of those who benefit from confusion and uncertainty rather than clarity and truth.

The trust crisis is compounded by the fact that many institutions and individuals have been slow to adapt to the new reality of AI-generated content. News organisations, academic institutions, and government agencies often lack clear policies for identifying, labelling, or responding to AI-generated content. This institutional uncertainty sends mixed signals to the public about how seriously to take the threat and what steps they should take to protect themselves.

The psychological impact of the trust crisis extends beyond rational calculation of information reliability. When people lose confidence in their ability to distinguish truth from falsehood, they may experience anxiety, paranoia, or learned helplessness. They may retreat into information bubbles where they only consume content from sources that confirm their existing beliefs, or they may become so overwhelmed by uncertainty that they disengage from public discourse entirely. Both responses undermine the informed public engagement that democratic societies require to function effectively.

The Explainability Imperative and Strategic Transparency

The demand for explainable AI has never been more urgent than in the context of information warfare. When AI systems are making decisions about what information to trust, what content to flag as suspicious, or how to respond to potential disinformation, the stakes are too high to accept black box decision-making. Democratic societies require transparency and accountability in the systems that shape public discourse, yet the most powerful AI technologies operate in ways that are fundamentally opaque to human understanding.

Explainable AI, often abbreviated as XAI, represents an attempt to bridge this gap by developing AI systems that can provide human-understandable explanations for their decisions. In the context of disinformation detection, this might mean an AI system that can not only identify a piece of content as potentially false but also explain which specific features led to that conclusion. Such explanations could help human fact-checkers understand and verify the system's reasoning, build trust in its conclusions, and identify potential biases or errors in its decision-making process.

However, the challenge of creating truly explainable AI systems is far more complex than it might initially appear. The most powerful AI systems derive their capabilities from their ability to identify subtle patterns and relationships in vast datasets—patterns that may be too complex for humans to understand even when explicitly described. An AI system might detect disinformation by recognising a combination of linguistic patterns, metadata signatures, and contextual clues that, when taken together, indicate artificial generation. But explaining this decision in human-understandable terms might require simplifications that lose crucial nuance or accuracy.

The trade-off between AI capability and explainability creates a fundamental dilemma for those developing systems to combat weaponised truth. More explainable systems may be less effective at detecting sophisticated disinformation, while more effective systems may be less trustworthy due to their opacity. This tension is particularly acute because the adversaries developing disinformation campaigns are under no obligation to make their systems explainable—they can use the most sophisticated black box technologies available, while defenders may be constrained by explainability requirements.

Current approaches to explainable AI in this domain focus on several different strategies. Some researchers are developing “post-hoc” explanation systems that attempt to reverse-engineer the reasoning of black box AI systems after they make decisions. Others are working on “interpretable by design” systems that sacrifice some capability for greater transparency. Still others are exploring “human-in-the-loop” approaches that combine AI analysis with human oversight and verification.

Each of these approaches has significant limitations. Post-hoc explanations may not accurately reflect the actual reasoning of the AI system, potentially creating false confidence in unreliable decisions. Interpretable by design systems may be insufficient to address the most sophisticated disinformation campaigns. Human-in-the-loop systems may be too slow to respond to rapidly evolving information warfare tactics or may introduce their own biases and limitations.

What's needed is a new design philosophy that goes beyond these traditional approaches—what we might call “strategic explainability.” Unlike post-hoc explanations that attempt to reverse-engineer opaque decisions, or interpretable-by-design systems that sacrifice capability for transparency, strategic explainability would build explanation capabilities into the fundamental architecture of AI systems from the ground up. This approach would recognise that in the context of information warfare, the ability to explain decisions is not just a nice-to-have feature but a core requirement for effectiveness.

Strategic explainability would differ from existing approaches in several key ways. First, it would prioritise explanations that are actionable rather than merely descriptive—providing not just information about why a decision was made but guidance about what humans should do with that information. Second, it would focus on explanations that are contextually appropriate, recognising that different stakeholders need different types of explanations for different purposes. Third, it would build in mechanisms for continuous learning and improvement, allowing explanation systems to evolve based on feedback from human users.

This new approach would also recognise that explainability is not just a technical challenge but a social and political one. The explanations provided by AI systems must be not only accurate and useful but also trustworthy and legitimate in the eyes of diverse stakeholders. This requires careful attention to issues of bias, fairness, and representation in both the AI systems themselves and the explanation mechanisms they employ.

The Automation Temptation and Moral Outsourcing

As the scale and speed of AI-powered disinformation continue to grow, there is an increasing temptation to respond with equally automated defensive systems. The logic is compelling: if human fact-checkers cannot keep pace with AI-generated false content, then perhaps AI-powered detection and response systems can level the playing field. However, this approach to automation carries significant risks that may be as dangerous as the problems it seeks to solve.

Fully automated content moderation systems, no matter how sophisticated, inevitably make errors in classification and context understanding. When these systems operate at scale without human oversight, small error rates can translate into thousands or millions of incorrect decisions. In the context of information warfare, these errors can have serious consequences for free speech, democratic discourse, and public trust. False positives can lead to the censorship of legitimate content, while false negatives can allow harmful disinformation to spread unchecked.

The temptation to automate defensive responses is particularly strong for technology platforms that host billions of pieces of content and cannot possibly review each one manually. However, automated systems struggle with the contextual nuance that is often crucial for distinguishing between legitimate and harmful content. A factual statement might be accurate in one context but misleading in another. A piece of satire might be obviously humorous to some audiences but convincing to others. A historical document might contain accurate information about past events but be used to spread false narratives about current situations.

Beyond these technical limitations lies a more fundamental concern: the ethical risk of moral outsourcing to machines. When humans delegate moral judgement to black-box detection systems, they risk severing their own accountability for the consequences of those decisions. This delegation of moral responsibility represents a profound shift in how societies make collective decisions about truth, falsehood, and acceptable discourse.

The problem of moral outsourcing becomes particularly acute when we consider that AI systems, no matter how sophisticated, lack the moral reasoning capabilities that humans possess. They can be trained to recognise patterns associated with harmful content, but they cannot understand the deeper ethical principles that should guide decisions about free speech, privacy, and democratic participation. When we automate these decisions, we risk reducing complex moral questions to simple technical problems, losing the nuance and context that human judgement provides.

This delegation of moral authority to machines also creates opportunities for those who control the systems to shape public discourse in ways that serve their interests rather than the public good. If a small number of technology companies control the AI systems that determine what information people see and trust, those companies effectively become the arbiters of truth for billions of people. This concentration of power over information flows represents a fundamental threat to democratic governance and pluralistic discourse.

The automation of defensive responses also creates the risk of adversarial exploitation. Bad actors can study automated systems to understand their decision-making patterns and develop content specifically designed to evade detection or trigger false positives. They can flood systems with borderline content designed to overwhelm human reviewers or force automated systems to make errors. They can even use the defensive systems themselves as weapons by manipulating them to censor legitimate content from their opponents.

The challenge is further complicated by the fact that different societies and cultures have different values and norms around free speech, privacy, and information control. Automated systems designed in one cultural context may make decisions that are inappropriate or harmful in other contexts. The global nature of digital platforms means that these automated decisions can affect people around the world, often without their consent or awareness.

The alternative to full automation is not necessarily manual human review, which is clearly insufficient for the scale of modern information systems. Instead, the most promising approaches involve human-AI collaboration, where automated systems handle initial screening and analysis while humans make final decisions about high-stakes content. These hybrid approaches can combine the speed and scale of AI systems with the contextual understanding and moral reasoning of human experts.

However, even these hybrid approaches must be designed carefully to avoid the trap of moral outsourcing. Human oversight must be meaningful rather than perfunctory, with clear accountability mechanisms and regular review of automated decisions. The humans in the loop must be properly trained, adequately resourced, and given the authority to override automated systems when necessary. Most importantly, the design of these systems must preserve human agency and moral responsibility rather than simply adding a human rubber stamp to automated decisions.

The Defensive Paradox

The development of AI-powered defences against disinformation creates a paradox that strikes at the heart of the entire enterprise. The same technologies that enable sophisticated disinformation campaigns also offer our best hope for detecting and countering them. This dual-use nature of AI technology means that advances in defensive capabilities inevitably also advance offensive possibilities, creating an escalating cycle where each improvement in defence enables corresponding improvements in attack.

This paradox is particularly evident in the development of detection systems. The most effective approaches to detecting AI-generated disinformation involve training AI systems on large datasets of both authentic and artificial content, teaching them to recognise the subtle patterns that distinguish between the two. However, this same training process also teaches the systems how to generate more convincing artificial content by learning which features detection systems look for and how to avoid them.

The result is that every advance in detection capability provides a roadmap for improving generation systems. Researchers developing better detection methods must publish their findings to advance the field, but these publications also serve as instruction manuals for those seeking to create more sophisticated disinformation. The open nature of AI research, which has been crucial to the field's rapid advancement, becomes a vulnerability when applied to adversarial applications.

This dynamic creates particular challenges for defensive research. Traditional cybersecurity follows a model where defenders share information about threats and vulnerabilities to improve collective security. In the realm of AI-powered disinformation, this sharing of defensive knowledge can directly enable more sophisticated attacks. Researchers must balance the benefits of open collaboration against the risks of enabling adversaries.

The defensive paradox also extends to the deployment of counter-disinformation systems. The most effective defensive systems might need to operate with the same speed and scale as the offensive systems they're designed to counter. This could mean deploying AI systems that generate counter-narratives, flood false information channels with authentic content, or automatically flag and remove suspected disinformation. However, these defensive systems could easily be repurposed for offensive operations, creating powerful tools for censorship or propaganda.

The challenge is compounded by the fact that the distinction between offensive and defensive operations is often unclear in information warfare. A system designed to counter foreign disinformation could be used to suppress legitimate domestic dissent. A tool for promoting accurate information could be used to amplify government propaganda. The same AI capabilities that protect democratic discourse could be used to undermine it.

The global nature of AI development exacerbates this paradox. While researchers in democratic countries may be constrained by ethical considerations and transparency requirements, their counterparts in authoritarian regimes face no such limitations. This creates an asymmetric situation where defensive research conducted openly can be exploited by offensive actors operating in secret, while defensive actors cannot benefit from insights into offensive capabilities.

The paradox is further complicated by the fact that the most sophisticated AI systems are increasingly developed by private companies rather than government agencies or academic institutions. These companies must balance commercial interests, ethical responsibilities, and national security considerations when deciding how to develop and deploy their technologies. The competitive pressures of the technology industry can create incentives to prioritise capability over safety, potentially accelerating the development of technologies that could be misused.

The Speed of Deception

One of the most transformative aspects of AI-powered disinformation is the speed at which it can be created, deployed, and adapted. Traditional disinformation campaigns required significant human resources and time to develop and coordinate. Today's AI systems can generate thousands of unique pieces of false content in minutes, distribute them across multiple platforms simultaneously, and adapt their messaging in real-time based on audience response.

This acceleration fundamentally changes the dynamics of information warfare. In the past, there was often a window of opportunity for fact-checkers, journalists, and other truth-seeking institutions to investigate and debunk false information before it gained widespread traction. Today, false narratives can achieve viral spread before human fact-checkers are even aware of their existence. By the time accurate information is available, the false narrative may have already shaped public opinion and moved on to new variations.

The speed advantage of AI-generated disinformation is particularly pronounced during breaking news events, when public attention is focused and emotions are heightened. AI systems can immediately generate false explanations for unfolding events, complete with convincing details and emotional appeals, while legitimate news organisations are still gathering facts and verifying sources. This creates a “first-mover advantage” for disinformation that can be difficult to overcome even with subsequent accurate reporting.

The rapid adaptation capabilities of AI systems create additional challenges for defenders. Traditional disinformation campaigns followed relatively predictable patterns, allowing defenders to develop specific countermeasures and responses. AI-powered campaigns can continuously evolve their tactics, testing different approaches and automatically optimising for maximum impact. They can respond to defensive measures in real-time, shifting to new platforms, changing their messaging, or adopting new techniques faster than human-operated defence systems can adapt.

This speed differential has profound implications for democratic institutions and processes. Elections, policy debates, and other democratic activities operate on human timescales, with deliberation, discussion, and consensus-building taking days, weeks, or months. AI-powered disinformation can intervene in these processes on much faster timescales, potentially disrupting democratic deliberation before it can occur. The result is a temporal mismatch between the speed of artificial manipulation and the pace of authentic democratic engagement.

The challenge is further complicated by the fact that human psychology is not well-adapted to processing information at the speeds that AI systems can generate it. People need time to think, discuss, and reflect on important issues, but AI-powered disinformation can overwhelm these natural processes with a flood of compelling but false information. The sheer volume and speed of artificially generated content can make it difficult for people to distinguish between authentic and artificial sources, even when they have the skills and motivation to do so.

The speed of AI-generated content also creates challenges for traditional media and information institutions. News organisations, fact-checking services, and academic researchers all operate on timescales that are measured in hours, days, or weeks rather than seconds or minutes. By the time these institutions can respond to false information with accurate reporting or analysis, the information landscape may have already shifted to new topics or narratives.

The International Dimension

The global nature of AI development and digital communication means that the challenge of weaponised truth cannot be addressed by any single nation acting alone. Disinformation campaigns originating in one country can instantly affect populations around the world, while the AI technologies that enable these campaigns are developed and deployed across multiple jurisdictions with different regulatory frameworks and values.

This international dimension creates significant challenges for coordinated response efforts. Different countries have vastly different approaches to regulating speech, privacy, and technology development. What one nation considers essential content moderation, another might view as unacceptable censorship. What one society sees as legitimate government oversight, another might perceive as authoritarian control. These differences in values and legal frameworks make it difficult to develop unified approaches to combating AI-powered disinformation.

The challenge is compounded by the fact that some of the most sophisticated disinformation campaigns are sponsored or supported by nation-states as part of their broader geopolitical strategies. These state-sponsored operations can draw on significant resources, technical expertise, and intelligence capabilities that far exceed what private actors or civil society organisations can deploy in response. They can also exploit diplomatic immunity and sovereignty principles to shield their operations from legal consequences.

The struggle over AI and information has become a central theatre in the U.S.-China superpower competition, with experts warning that the United States is “not prepared to defend or compete in the AI era.” This geopolitical dimension transforms the challenge of weaponised truth from a technical problem into a matter of national security. A partial technological separation between the U.S. and China, particularly in AI, is already well underway, creating parallel development ecosystems with different standards, values, and objectives.

This technological decoupling has significant implications for global efforts to combat disinformation. If the world's two largest economies develop separate AI ecosystems with different approaches to content moderation, fact-checking, and information verification, it becomes much more difficult to establish global standards or coordinate responses to cross-border disinformation campaigns. The result could be a fragmented information environment where different regions of the world operate under fundamentally different assumptions about truth and falsehood.

The international AI research community faces particular challenges in balancing open collaboration with security concerns. The tradition of open research and publication that has driven rapid advances in AI also makes it easier for bad actors to access cutting-edge techniques and technologies. Researchers developing defensive capabilities must navigate the tension between sharing knowledge that could help protect democratic societies and withholding information that could be used to develop more sophisticated attacks.

International cooperation on AI governance has made some progress through forums like the Partnership on AI, the Global Partnership on AI, and various UN initiatives. However, these efforts have focused primarily on broad principles and voluntary standards rather than binding commitments or enforcement mechanisms. The pace of technological change often outstrips the ability of international institutions to develop and implement coordinated responses.

The private sector plays a crucial role in this international dimension, as many of the most important AI technologies are developed by multinational corporations that operate across multiple jurisdictions. These companies must navigate different regulatory requirements, cultural expectations, and political pressures while making decisions that affect global information flows. The concentration of AI development in a relatively small number of large companies creates both opportunities and risks for coordinated response efforts.

Expert consensus on the future of the information environment remains fractured, with researchers “evenly split” on whether technological and societal solutions can overcome the rise of false narratives, or if the problem will worsen. This lack of consensus reflects the genuine uncertainty about how these technologies will evolve and how societies will adapt to them. It also highlights the need for continued research, experimentation, and international dialogue about how to address these challenges.

Looking Forward: The Path to Resilience

The challenges posed by AI-powered disinformation and weaponised truth are unlikely to be solved through any single technological breakthrough or policy intervention. Instead, building resilience against these threats will require sustained effort across multiple domains, from technical research and policy development to education and social change. The goal should not be to eliminate all false information—an impossible and potentially dangerous objective—but to build societies that are more resistant to manipulation and better able to distinguish truth from falsehood.

Technical solutions will undoubtedly play an important role in this effort. Continued research into explainable AI, adversarial robustness, and human-AI collaboration could yield tools that are more effective and trustworthy than current approaches. Advances in cryptographic authentication, blockchain verification, and other technical approaches to content provenance could make it easier to verify the authenticity of digital information. Improvements in AI safety and alignment research could reduce the risk that defensive systems will be misused or corrupted.

However, technical solutions alone will be insufficient without corresponding changes in policy, institutions, and social norms. Governments need to develop more sophisticated approaches to regulating AI development and deployment while preserving innovation and free expression. Educational institutions need to help people develop better critical thinking skills and digital literacy. News organisations and other information intermediaries need to adapt their practices to the new reality of AI-generated content.

The development of strategic explainability represents a particularly promising avenue for technical progress. By building explanation capabilities into the fundamental architecture of AI systems from the ground up, researchers could create tools that are both more effective at detecting disinformation and more trustworthy to human users. This approach would recognise that in the context of information warfare, the ability to explain decisions is not just a desirable feature but a core requirement for effectiveness.

The challenge of moral outsourcing to machines must also be addressed through careful system design and governance structures. Human oversight of AI systems must be meaningful rather than perfunctory, with clear accountability mechanisms and regular review of automated decisions. The humans in the loop must be properly trained, adequately resourced, and given the authority to override automated systems when necessary. Most importantly, the design of these systems must preserve human agency and moral responsibility rather than simply adding a human rubber stamp to automated decisions.

The international community must also develop new mechanisms for cooperation and coordination in addressing these challenges. This could include new treaties or agreements governing the use of AI in information warfare, international standards for AI development and deployment, and cooperative mechanisms for sharing threat intelligence and defensive technologies. Such cooperation will require overcoming significant political and cultural differences, but the alternative—a fragmented response that allows bad actors to exploit regulatory arbitrage—is likely to be worse.

The ongoing technological decoupling between major powers creates additional challenges for international cooperation, but it also creates opportunities for like-minded nations to develop shared approaches to AI governance and information security. Democratic countries could work together to establish common standards for AI development, create shared defensive capabilities, and coordinate responses to disinformation campaigns. Such cooperation would need to be flexible enough to accommodate different national values and legal frameworks while still providing effective collective defence.

Perhaps most importantly, societies need to develop greater resilience at the human level. This means not just better education and critical thinking skills, but also stronger social institutions, healthier democratic norms, and more robust systems for collective truth-seeking. It means building communities that value truth over tribal loyalty and that have the patience and wisdom to engage in thoughtful deliberation rather than rushing to judgment based on the latest viral content.

The psychological and social dimensions of the challenge require particular attention. People need to develop better understanding of how their own cognitive biases can be exploited, how to evaluate information sources critically, and how to maintain healthy scepticism without falling into cynicism or paranoia. Communities need to develop norms and practices that support constructive dialogue across different viewpoints and that resist the polarisation that makes disinformation campaigns more effective.

Educational institutions have a crucial role to play in this effort, but traditional approaches to media literacy may be insufficient for the challenges posed by AI-generated content. New curricula need to help people understand not just how to evaluate information sources but how to navigate an information environment where the traditional markers of credibility may no longer be reliable. This education must be ongoing rather than one-time, as the technologies and tactics of information warfare continue to evolve.

The stakes in this effort could not be higher. The ability to distinguish truth from falsehood, to engage in rational public discourse, and to make collective decisions based on accurate information are fundamental requirements for democratic society. If we fail to address the challenges posed by weaponised truth and AI-powered disinformation, we risk not just the spread of false information but the erosion of the epistemological foundations that make democratic governance possible.

The path forward will not be easy, and there are no guarantees of success. The technologies that enable weaponised truth are powerful and rapidly evolving, while the human vulnerabilities they exploit are deeply rooted in our psychology and social behaviour. But the same creativity, collaboration, and commitment to truth that have driven human progress throughout history can be brought to bear on these challenges. The question is whether we will act quickly and decisively enough to build the defences we need before the weapons become too powerful to counter.

The future of truth in the digital age is not predetermined. It will be shaped by the choices we make today about how to develop, deploy, and govern AI technologies. By acknowledging the challenges honestly, working together across traditional boundaries, and maintaining our commitment to truth and democratic values, we can build a future where these powerful technologies serve human flourishing rather than undermining it. The stakes are too high, and the potential too great, for any other outcome to be acceptable.

---

References and Further Information

Primary Sources:

Understanding Russian Disinformation and How the Joint Force Can Counter It – U.S. Army War College Publications, publications.armywarcollege.edu

Future Shock: Generative AI and the International AI Policy and Governance Landscape – Harvard Data Science Review, hdsr.mitpress.mit.edu

The Future of Truth and Misinformation Online – Pew Research Center, www.pewresearch.org

U.S.-China Technological “Decoupling”: A Strategy and Policy Framework – Carnegie Endowment for International Peace, carnegieendowment.org

Setting the Future of Digital and Social Media Marketing Research: Perspectives and Research Propositions – Science Direct, www.sciencedirect.com

Problems with Autonomous Weapons – Campaign to Stop Killer Robots, www.stopkillerrobots.org

Countering Disinformation Effectively: An Evidence-Based Policy Guide – Carnegie Endowment for International Peace, carnegieendowment.org

Additional Research Areas:

Partnership on AI – partnershiponai.org Global Partnership on AI – gpai.ai MIT Center for Collective Intelligence – cci.mit.edu Stanford Human-Centered AI Institute – hai.stanford.edu Oxford Internet Institute – oii.ox.ac.uk Berkman Klein Center for Internet & Society, Harvard University – cyber.harvard.edu

---

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0000-0002-0156-9795 Email: tim@smarterarticles.co.uk

Discuss...