The COVID-19 pandemic left no doubt that biosecurity emergencies are not purely biological events; they are also communication and trust crises. Throughout the global health emergency, public health advice, scientific research, political rhetoric, and digital media became deeply interdependent, meaning that the quality of information governance was as important as the quality of laboratory science. The World Health Organization captured this dimension when it described the pandemic as being accompanied by an “infodemic”—an overabundance of information, some accurate and some false, that made it difficult for people to find reliable guidance and take protective action. This was not merely a metaphor: in the digital age, health emergencies unfold simultaneously in the body, in the media ecosystem, and in the public psyche, and the volume of information grew faster than the ability of people or institutions to filter it. The challenge has not disappeared with the end of the acute phase of COVID-19. Artificial intelligence and open-source intelligence are now being used to detect and monitor biological threats, with AI systems scanning social media platforms, online news outlets, blogs, web forums, and other publicly available sources to identify outbreak signals earlier than traditional surveillance systems might. These tools hold considerable promise for improving early warning and situational awareness. Yet they also create a novel governance problem: the same platforms that provide valuable open-source signals are the platforms through which misinformation, disinformation, and malicious influence operations spread. Governments and public health agencies had to make decisions in real time, often before the science was settled. As new data emerged, recommendations changed, creating an appearance of inconsistency. Meanwhile, social media amplified everything—credible research, provisional findings, and outright falsehoods—with little to distinguish among them. In such an environment, public trust became a casualty of information overload. The critical question, as articulated by Shravishtha Ajaykumar, an Associate Fellow at the Observer Research Foundation’s Centre for Security, Strategy, and Technology, is how to distinguish between legitimate scientific communication, false information, and honest scientific dissent when the infrastructure used for crisis management is also the infrastructure used to contaminate the information environment.
The experience of COVID-19 illustrated these dynamics vividly. At the outset of the pandemic, there was no consensus on many basic features of the virus, including its modes of transmission. Official advice initially stressed the importance of respiratory droplets and contaminated surfaces, while a number of aerosol scientists argued that airborne transmission was a more significant route than official guidance acknowledged. This was not a case of misinformation; it was a legitimate scientific disagreement that reflected genuine uncertainty in the early stage of knowledge accumulation. As evidence accumulated, public health recommendations changed—on mask use, indoor ventilation, vaccines, and social distancing—and updated guidance increasingly recognized the role of airborne transmission. But to many members of the public, the changing advice looked like confusion or incompetence. The information environment amplified this perception. Unverified claims, miracle cures, anti-vaccine narratives, and conspiracy theories about the virus’s origin travelled alongside credible research and official announcements. Political disputes over lockdowns, school closures, and mandates further blurred the boundary between science and policy. Consequently, the pandemic demonstrated that scientific disagreement is not inherently dangerous; indeed, it is a normal and necessary part of science. The danger arises when disagreement is presented as evidence of failure, when uncertainty is exploited to undermine trust, and when false information is allowed to circulate with the same authority as rigorous research. The distinction between healthy scientific contest and harmful misinformation must therefore be understood not as a binary but as a spectrum that changes as knowledge evolves. Part of the reason the pandemic was so hard to communicate was that official bodies often failed to acknowledge the difference between a settled finding and a provisional one. When guidance was presented with more certainty than the evidence allowed, any later correction became a proof of untrustworthiness. Conversely, when officials expressed uncertainty without explaining the process by which it would be resolved, they left a vacuum that misinformation quickly filled. The middle ground—transparently communicating what is known, what is not known, and how knowledge is expected to change—was often the hardest to achieve in real time. The pandemic showed that communication cannot be separated from epidemiology; every technical decision was also a communication decision.
Into this already turbulent information landscape has come a new generation of AI-based biosecurity tools. These systems draw on open-source intelligence—OSINT—and use machine learning and natural language processing to monitor vast streams of data from social media, news websites, blogs, forums, and other publicly available sources. The goal is to detect emerging disease signals, sometimes before they are recognized by official surveillance systems. Such early warning systems could substantially improve pandemic preparedness, especially in regions with weak health data infrastructure. But they also introduce serious risks. During the pandemic, false narratives about COVID-19’s origins, vaccines, and treatments spread at digital speed. If an AI system ingests this kind of content without discrimination, it may treat malicious falsehoods as genuine signals, producing false alarms that overwhelm public health agencies. It may also miss real outbreaks because its attention is diverted by noisy or manufactured data. Worse, state or non-state actors might deliberately seed information to manipulate the systems, using fake posts and bot networks to create the impression of an outbreak where none exists, or to obscure one that does. Ajaykumar argues that the value of AI in biosecurity will therefore depend not only on more sophisticated algorithms but on the quality, provenance, and verification of open-source information. Publicly available data cannot be considered equally reliable. A tweet from an anonymous account is not equivalent to a peer-reviewed study; a blog post is not equivalent to a government disease surveillance report. Standards for evaluating sources, assigning confidence levels, and cross-checking information against official data are essential if AI-assisted surveillance is to provide a trustworthy basis for decisions. Indeed, the more such systems rely on open-source data, the more vulnerable they become to the information disorder that characterised the pandemic. If a machine cannot distinguish between a scientific preprint and a conspiracy theory, its output is not an early warning; it is a reflection of the chaos of the internet. To be clear, AI systems can be trained to filter sources, but the same datasets that contain useful local signals also contain deliberate manipulation. There is no technical fix that removes the need for judgment. Human analysts must be able to see how an algorithm reached its conclusions and must be able to override them when the underlying information appears dubious. Transparency about data sources and model assumptions should be built into the design of such systems, not bolted on later.
The problem of distinguishing misinformation from legitimate dissent has already resurfaced in later outbreaks. The 2022 mpox outbreak, for example, showed that public health communication is never just about the science. Because the virus initially spread through specific sexual networks, officials had to convey risk to those communities without fuelling stigma or discouraging testing and treatment. Even accurate information could be harmful if it was framed in ways that associated a disease with a particular identity or behaviour, or if it ignored the social context in which people live. In such circumstances, trust is not simply a matter of factual accuracy; it is also a matter of respect, framing, and inclusivity. H5N1 avian influenza poses a different communication challenge. Scientists and public health agencies must walk a fine line between preparing for a potential pandemic and avoiding both panic and complacency. Human cases of H5N1 have been reported, but sustained human-to-human transmission has not yet occurred. Public statements need to make clear that a biological hazard exists without implying that a worst-case scenario is imminent. If authorities overstate the risk, they may provoke fear and erode credibility when the expected pandemic does not materialize; if they understate it, they may be accused of suppressing information if the virus changes. Genomic surveillance deepens these dilemmas. During COVID-19, the rapid identification of variants such as Alpha, Delta, and Omicron allowed governments to adjust vaccines and policies, but it also turned genetic findings into objects of public speculation and misinformation. In an AI-driven early warning environment, inaccurate, manipulated, or deliberately leaked genomic or epidemiological data could influence threat assessments and international reporting. The underlying issue is that scientific knowledge will necessarily evolve during an outbreak. No one can promise complete certainty at the beginning of an unknown disease. The question is whether governance systems can accommodate uncertainty without losing public trust. This will require explicit standards for evaluating evidence, transparent communication of uncertainty, and clear procedures for revising recommendations when new data arrive. At the same time, systems that aggregate data must be able to separate a credentialed scientific critique from an ideological attack on public health institutions. Both may look alike to an algorithm. Indeed, the boundary between scientific communication, misinformation, disinformation, and dissent is now an operational question for biosecurity systems, not merely an ethical one.
Whatever the technical safeguards, the deeper lesson of the pandemic is that trust in biosecurity emergencies must be built in advance and distributed across many institutions. It cannot be concentrated in a single politician, ministry, or international body. The World Health Organization will remain important as the coordinator of global health emergency response, setting norms, sharing surveillance information, and providing technical assistance to countries. But its credibility depends on member states reporting outbreaks transparently and on the WHO itself being able to communicate honestly about what is known and what is not. National public health agencies, in turn, are often the most visible sources of guidance during an emergency. They need sufficient resources, scientific independence, and a clear separation from partisan politics. If every public health measure is perceived as a political choice, then trust will collapse as soon as the government changes. Independent academic institutions, scientific journals, and researchers also play a vital role. They can validate or question official claims, provide an evidence base for decisions, and correct the record when errors are identified. The media, too, are part of the trust ecosystem; responsible reporting can channel expert knowledge to the public, while sensationalism or false balance can amplify uncertainty. This ecosystem requires preparation. Communication strategies should be integrated into public health emergency plans before an outbreak, not added after the fact. Governments should publish the data, models, and expert advice underlying their decisions, including the limitations and uncertainties. Public education should aim not to produce uniform agreement but to improve scientific literacy, so that people understand why guidance might change as evidence accumulates. Finally, because pathogens do not recognise borders, international mechanisms for data sharing and joint response are essential. The proposed WHO Pandemic Agreement, despite its difficult negotiations, reflects a widespread acknowledgement that no country can manage a biosecurity threat alone and that information sharing must be governed by rules agreed in advance. Trust that is built only during a crisis is typically trust that is already lost. People notice when messages shift, and they are more forgiving if they understand why. Officials should therefore rehearse their communication responses, engage with communities who are likely to be affected, and create channels for feedback and correction. Building trust is not merely a public relations exercise; it is a governance function as important as stockpiling ventilators or sequencing genomes.
AI-assisted early warning systems require a similarly deliberate approach to governance. Ajaykumar argues that standards must be established for how open-source data are collected, weighted, and verified. Not all data are equal: a signal from a Twitter account, a blog, or an anonymous forum should not be treated with the same confidence as a report from a credible public health agency or a peer-reviewed study. Each piece of information should carry provenance—an explicit record of where it came from, who produced it, and how it was obtained—so that analysts can judge its reliability. AI outputs should be cross-checked against authoritative disease surveillance data, and confidence levels should be assigned to early warnings based on the strength and diversity of the evidence. Human oversight should remain central, particularly when a system’s output might influence public health policy or international reporting. Regulation also has a role. Ajaykumar calls for rules that promote transparency about data sources, require validation of open-source data, audit AI systems for bias and vulnerability to misinformation, and ensure accountability for decisions made with the help of AI. Standards should also address the accountability of the organisations that deploy AI systems. If an early-warning system produces a false alarm that disrupts trade or travel, there must be a process for review and correction. If an algorithm is shown to be biased against certain countries or communities, those affected need a way to challenge its outputs. International cooperation will be necessary to agree on these standards, since outbreaks and the data generated around them are global. Technical fixes alone will not be enough. Without such safeguards, the use of AI in biosecurity could amplify the very information disorder it is meant to help manage. The conclusion of the analysis is clear: the line between misinformation and scientific dissent is among the most difficult challenges in biosecurity governance today. The COVID-19 pandemic showed that managing information is just as critical as managing the pathogen. Successful governance does not aim to eliminate disagreement or to impose a single official truth. Science depends on debate, and novel outbreaks inevitably contain uncertainty. The aim, instead, is to ensure that disagreements are resolved through evidence, that uncertainty is communicated honestly, and that the systems used to detect threats are not corrupted by falsehoods. That means treating scientific communication, data validation, and AI governance as core components of pandemic preparedness, not as afterthoughts. By doing so, societies can reap the benefits of artificial intelligence and open-source intelligence while protecting against the dangers of misinformation, disinformation, and the erosion of public trust.
