Can AI Surveillance learn from Medieval Surveillance?
David Lyon
David Lyon is a sociologist who directed the Surveillance Studies Centre at Queen's University in Kingston, Ontario.
I want to make some comments about AI surveillance. Though I am not an AI expert, I do recognize that both the input—the “training” regime—of AI and the output—I’ll mention e.g. facial recognition for public safety—of AI are surveillant, often in very negative ways. Whatever the specific area of surveillance, from consumer tracking to policing, and from health and welfare to workplace surveillance, AI is invoked, implicated and impossible to ignore.
And while I care about matters such as privacy, data justice, and so on, I do so, not for their own sake, but because they relate to a larger concern—human flourishing. Recently, in my search for coherent and creative ways of confronting contemporary surveillance, I’ve been pushing the historical envelope, to examine not only modern but also pre-modern modes of surveillance—even when they had no inclusive word for these. I’ll conclude with some remarks based on that research.
Personally, I have to acknowledge the benefits of surveillance. For example, as a 3 or 4 year old, living in the Belgian Congo, my parents could see that I had been struck down by mosquito-borne malaria, and that I would likely die without quinine. The clinic was out of quinine and everyone was waiting for the steamship to bring some up the river. It came, but only just in time. Much more recently, in my mid-sixties, I had just started the swim section of a triathlon when I discovered that I could no longer swim. Seen by the guard, in his kayak, I told him that I couldn’t swim any more, and he called the rescue boat to pick me up. In the hospital emergency department, the physician declared that I’d had a heart attack (actually caused, again, by a tiny insect—a tick). Without surveillance, I‘d not be here today.
How do we understand “surveillance”? For a long time, I thought of surveillance as
the focused, systematic and routine attention to personal details for the purposes of influence, management, protection or direction
But over recent years, especially since social media became significant, and surveillance became all-inclusive, and ubiquitous, I now add that the experience of surveillance should also be considered in any full definition.
At every turn, artificial intelligence (AI) systems continue to be hyped in 2025, while significant risks continue to be posed. One major set of risks is the surveillance use of AI, that many think can be reduced to a “data privacy” problem. But it’s much, much more than just privacy and this can no longer be ignored. Indeed, as I say, AI must itself be considered a surveillance technology due to its ability to collect, analyze, interpret and use vast amounts of data. The Panglossian pundits praise the possibilities of AI—some of which, I grant, may be genuine—but the real need is for a sober assessment of its risks, regarding everybody’s human rights, but more broadly, its contribution to our human being; our “humanity.”
The Trump regime offers some egregious examples of reckless faith in AI—such as the proposal for a decade-long moratorium on state-level AI regulation (showing it’s clearly not just a “corporate takeover” but an “AI coup”—as Naomi Klein puts it—led by Silicon Valley). Yet if one listens to international leaders (UK, FR, AU, CAN) they all harp on about the opportunities and advantages of the shift to AI and its role in competitiveness etc. As Emmanuel Macron announced in 2025, "…we will adopt the Notre Dame de Paris strategy" (France rebuilt the landmark cathedral within five years of its devastation in a 2019 fire). "We showed the rest of the world that when we commit to a clear timeline we can deliver…" "You decide, you streamline all the procedures, somebody is in charge," he added, saying the scheme would apply to data centres, authorisations to bring AI products to market and business "attractiveness." The rapid developments in the training and use of AI have raised concerns about user consent, the ethical use of personal data and privacy rights in general. But there is also a host of other pressing issues, including, importantly, what about the clear connections between AI and surveillance and priorities such as care, justice, relationality and accountability?
AI and Facial Recognition
First, it's commonly known that AI models may inherit overlooked biases. For example, a Canadian-American MIT PhD student in computing, Joy Buolamwini, discovered from her research that the datasets produced for facial recognition—that is, those present in the training data—were markedly skewed. The reliability of the “recognition” varied strongly by gender and colour.
She spoke on the radio in Canada in May 2025. She now talks not only of the male gaze, the white gaze or the postcolonial gaze, but also of the “coded gaze.” It is no less power-laden than the others. She asks: “Who has the power to shape the priorities, the preferences—and also at times, maybe not intentionally—the prejudices that are embedded into technology?” As Buolamwini says, such can lead “…to things like false arrests, non-consensual deep fakes as well for explicit imagery. And it impacts everybody, especially when you have companies like Clearview AI, that has scraped billions of photos courtesy of social media platforms. Not that we gave them permission, but this is what they've done”
"So as we think about where we are in this stage of AI development…” she goes on, “I oftentimes think of the excoded—the excoded represents anyone who's been condemned, convicted, exploited, otherwise harmed by AI systems." This is clearly demonstrated by efforts to use AI-based facial-recognition for “Public safety” by police and security authorities.
Let me say something about facial recognition in a public safety context: AI facilitates the implementation of facial recognition technology which can identify and track individuals based on their facial features, even in the real world, and is used by police and security forces. This technology is already being used in public spaces, such as railway stations or airports, and by law enforcement agencies. The widespread use of facial recognition raises concerns about the constant monitoring of individuals. How did it evolve?
CCTV cameras started to appear in London UK in 1953 and proliferated in the 1990s—such that this city was for a while more densely covered than any in the world. In the same decade, facial recognition was trialled but abandoned due to the inadequate technology. However, once superior technology was available, new trials occurred, run by the Metropolitan Police, between 2016-2019. Independent academic researchers were invited to monitor and report on the trials
The London Metropolitan Police—the “Met”—is responsible for security at major public events, from King Charles’ coronation (TNW 2023)—that included the use of Live Facial Recognition due to anti-monarchy demonstrations—to large soccer games. But Fussey and Murray—in 2019—found that the Met system of “live” FRT was only reliable in 19% of cases, and that people were thus misidentified. Scanning millions of faces each year, facial recognition technology (FRT) has become one of today's fastest growing and most controversial AI-driven surveillance technologies. Fussey and Murray report on this in several published articles, as well as in their book that appears this summer
Based on their almost unprecedented ethnographic access to police FRT deployments, negotiated by Pete Fussey, the book on Facial Recognition Surveillance delves into the profound impact of FRT on policing practices, surveillance capabilities, and human rights protections. It reveals how this technology shapes, and is also shaped by, the complex environments in which it is deployed, dramatically reshaping police-citizen interactions in actual use. Fussey and Murray expose the selective scientific and legal narratives that justify the expansion of AI-driven surveillance. They draw on cutting-edge human rights theory--developed by Daragh Murray especially—to propose a due diligence framework tailored to police use of FRT and introduces the concept of 'compound human rights harm' to capture the multifaceted impact of surveillance. I commend their work to you.
Now, so-called smart cameras are frequently brought out for sporting and athletic events, predominantly at the Olympics—such as in Paris, France in 2024
France’s CNIL (Commission Nationale de l’Informatique et des Libertés) had clearly drawn the line at Live Facial Recognition or its equivalents. Today, however, Morocco, in preparation for the 2025-6 Africa Cup of Nations, and the co-hosted FIFA World Cup, AI-driven facial recognition and behaviour-tracking technologies are being bought (for US$10.8M). Many fear that the result will be to entrench state control and undermine basic rights to privacy, free speech and freedom of assembly
All-too-often, AI-based surveillance is seen as something that should be developed for contexts such as these—despite the poor track record of such experiments to date. As Isadora Borges Monroy
ICE foreigner deportation program in the USA.
Trump’s deportation regime involves intensified surveillance as the ICE-based data is raided for “evidence” of unwanted “foreign” nationals—even those with completely legal work permits, or have been accepted in the US for some time. Trump has promised to deport up to 21 million “illegal” immigrants. Because undocumented immigrants use the services—health, schooling—these are targets. By sending ICE agents to these sites they are deprived of rights and subjected to terrifying experiences (p2). This is apparently OK, for Trump sees these non-white immigrants as “animals”… “poisoning the blood of our country.” ICE agents are also likely to be seen on university campuses, seeking out international students.
“Another locus of immigrant surveillance is the US border. This includes ramping up walls and barriers; cameras, sensors, and drones; and personnel, including the possibility of sending 250,000 troops to patrol the US-Mexico border (ACLU 2024)”
Digital rights advocates remind of the tendency of AI tools to “hallucinate,” giving false information, and legal scholars worry that civil rights violations and abuses are increasing. Jonathan Guerrero e.g., a US citizen arrested by ICE in Philadelphia, and Jensy Machado, another US citizen arrested at gunpoint while driving to work in Virginia, are examples offered by Monahan. They were later released. Unfortunately, the current administration has shown no interest in accuracy, so much as evidence that they have “caught” illegals and are dealing with them. AI is being weaponized in destroying government programs—e.g. Elon Musk using Microsoft’s AI Azure program to find whistleblowers or eliminate programs for marginal populations. Surveillance systems target any and all US residents, whether or not they are citizens. Aggregated data is a problem—e.g. ICE has access to drivers’ licence data of 3 in 4 US adults and data-intensive tools can aggregate data points to implicate others in households, workplaces, or any other life-group. Aggregated data is used by ICE to determine who should be detained, released, or the terms of their surveillance.
“The DHS can also use ‘287(g) agreements’ (from a Trump executive order) to allow local law enforcement to act as federal immigration officers, giving them access to all ICE AI tools. This means thousands more ‘immigration agents’ handle personal data that may implicate people to be hunted”
From Artificial Intelligence to ‘Critical Intelligence’
AI has very quickly become a major issue in everyday social relations, in a variety of contexts, why we should be thoroughly skeptical of the claims made for AI. Yet I am not denying that AI can be harnessed to “human-centred” projects such as the care of Alzheimer’s sufferers, or the imaginative work of Maori efforts to reclaim their language using AI speech recognition, discussed by Karen Hao in The Empire of AI (2025). I referred earlier to the work of Shoshana Zuboff in her well-known and very popular treatise on “surveillance capitalism.” In claiming that we should seek not merely “artificial” but also critical intelligence,
I am arguing that we should place AI in the context of the most engaging and relevant analyses, that look, not only to the global north, and the Silicon Valley companies, but also to the global south.
Data Colonialism and Critical Intelligence
With regard to the global south, let me just mention here one relevant study: Nick Couldry and Ulises Mejias, on "data colonialism." The term refers to the global north as well as the global south, and of course, this also invites nuanced analyses referring to each. They demonstrate the similarities between what they call the "data grab" of high-tech corporations—described in its early phases by Zuboff—and the attitudes and actions of classic colonialism, especially in its C19th and C20th forms. Instead of "grabbing" the land, resources and labour of colonized countries through enforced extraction, what is now grabbed is data, which of course includes much that refers to individuals and to classes of societal members. And rather than claims about "terra nullius" or "nobody's land," the claim, to quote Google, is that data is as "free as the sunlight." It belongs to no one and therefore may be extracted at will with a clear conscience.
AI algorithms can analyze patterns of behavior, both in the real world and in online spaces. This includes monitoring social media activities, online searches and communication patterns. Through the massive use of personal data to train AI systems, this technology becomes much like a surveillance system that can "know" what people are thinking and can predict what they are going to like, dislike or what they might do in a given context. Or at least, this is what is often claimed. In the realm of public security, it's suspicious or at least "abnormal" movement or action in an Olympic crowd, a dubious ID at the airport, or, if you wish to engage in AI assisted security yourself, you can use "Smart Video Search" an AI tool offered with your "Ring" doorbell to discover more about the unknown person near your doorstep earlier in the day, or even whose dog fouled your front path.
But needless to say, AI is all-too-frequently driven by economic criteria or political power or both together. What is needed is Critical Intelligence; a sense of appropriate and humanly beneficial technology. In the early days of computing, one heard clear warnings by leading figures—such as British pioneer Alan Turing or US computer scientist Joseph Weizenbaum—about the risks of computing, if certain protocols were not maintained. In Canada, leading figures in computing—e.g. Kelly Gotlieb at University of Toronto ("father of Canadian computing")—were very clear about the moral and ethical questions surrounding computing. That's why the SSC at Queen's, Canada, invited him to the first ever "Surveillance Studies" research workshop in 1993. He cared about the ethics of surveillance, arguing that "privacy" could not bear the weight of the issues arising in the field. His colleague, Ron Ragsdale, once wrote a book contrasting the "possible" and the "permissible" in computing. The first does not entail the second.
Around today's world, there are highly qualified voices speaking out against the unregulated developments in AI, demanding that clear guardrails and limits be set on AI innovations. I believe that surveillance studies scholars should be examining the work of AI professionals, especially those who sound the alarm, rather than those who are simply aiming at being the "first" to market some new development for surveillance purposes. Such professionals include people like Karen Hao, who started out as a Silicon Valley engineer,
"Autocratic governments are already using AI and social media to solidify their internal propaganda and control dissent (including the use of both internet surveillance and visual surveillance through face recognition). There is therefore a risk that AI, especially AGI, could help autocrats stay in power and increase their dominance, even bring about an autocratic world government."
Can We Plan for Livable Futures?
We already live in a world of big data, and the expansion of computing power through AI is drastically changing how people—not just "privacy"—are protected, or not. For people, and the planet, to flourish, freely, with reliable, accountable leadership, AI will have to be developed in very different ways from the current "data grab" of monopoly corporations and power-hungry autocratic government, dependent on the "AI Empire." AI assuredly is the latest technology presenting new challenges, not only to consumers, businesses and regulators but across the spectrum of human life. Appealing to companies to apply best practices for data privacy compliance to build user trust might be a starting point. One may think that the "best practices" on a government level are currently those of the European General Data Protection Regulation, as they attempt to grapple with the massive challenges thrown up by Big Data and AI surveillance. But there's much room for everyday practices of dissent and re-imagining AI through democratically contesting the technology—as Hao argues.
This won't happen at most AI-centred events, that tend to be only for "experts," either technology or financial and other corporate representatives. Yet their impacts are felt by everyone including people with disabilities, women, Indigenous (even when land acknowledgements are made!), non-human, immigrants, LGBTQ, YP, people in poverty, not to mention the global AI workforce. Also police, security, and military (where AI development is currently rampant) should also be present to explain and justify their use of AI. Petra Molnar, for example, has traced historical and current tech experimentation in contested spaces such as war zones, prisons and migrant-favoured borders.
Private-security companies are a potentially ripe group for testing dangerous tech.
People and groups who may be more critical or skeptical about AI claims are disproportionately missing from AI policy conferences, including the biggest, such as the 2023 world AI summits in Montreal and Amsterdam, the 2023 U.S. summits in Las Vegas and New York, and the 2023 U.K. safety summit in London. Smaller conferences tend to focus even more on a small number of topics and groups. It's up to other kinds of democratically-oriented groups to work towards a different AI—and they do exist!
Ancient Light on a Present Problem?
Personally, I regard Yoshua Bengio as a pioneer in his work, not just for his advanced technical knowledge, but because he refuses to be subverted by the machine. His is a human-centred approach that warns against the speed of AI development and its potentially devastating possibilities for human life on planet earth—especially social and political human life. I'm now exploring how far this relates to my current research project. After several decades of analysing modern surveillance—including, now, AI-assisted surveillance—I'm currently exploring surveillance in pre-modern, including early medieval times in Europe.
While examples of these are found in the present, my current research takes us back to early medieval times and to the early precedents of what we now call 'surveillance'—based then around the idea of the "eye of God." It appears in the C5th writings of Augustine, for example, and later on (C15th) in figures such as Nicholas of Cusa. This was a very human-centred approach—seen in various contexts, from monasteries to early legal structures, and from there to literature and to farm and town security. Key ideas central to this were that "eye of God" surveillance was relational (not based on the modern notion of the individual), caring (Michel Foucault discussed this at length, arguing that it was what he describes as a "power of care"),
I should observe that this foray into medieval history differs from some other works, such as Jensen's work on "The Medieval Internet."
Let me just comment briefly on the four motifs that I mentioned. The "eye of God" is relational rather than having a primarily "individual" focus. In May (2025) I was in Mexico at FLACSO and a former grad student took us to an old colonial church building in Metepec. As soon as we entered, I saw the gold-coloured depictions of the "eye" within its triangular frame (like the US $1 bill). The triangle is a reference to the Christian belief that God is not singular, but three-persons-in-one. God is social. The "eye" represents that sociality, and also assumes plurality in what the "eye" sees—human beings in relation. This immediately contrasts with the individualism of modernity, expressed in our surveillance systems, in the attitudes of many who interact with the world of so-called big data and algorithms, and in many of the laws that attempt to curtail the most egregious forms of surveillance.
Second, Michel Foucault discussed the matter of a "pastoral" emphasis in medieval surveillance, where the figure of the "shepherd" watching over sheep is central.
A third motif of early medieval surveillance was that it was justice-seeking, which is an extension of the "care" motif. Because the shepherd had a special eye for the vulnerable and indeed singled them out for special treatment, it's fair to say that justice was being sought. Note here that the biblical prophetic pronouncements are witness to the fact that YHWH "watches" for justice on the earth, and expects justice to be done. Indeed, the key motif running through the whole Bible is the "exodus" of Hebrew slaves from Egypt, that was prompted by the eye-of-God, which, we're told, saw and observed the suffering of the foreign workers: "I have surely seen the affliction of my people who are in Egypt, and have heard their cry because of their taskmasters." (Ex 3:7-8)
Lastly, medieval surveillance started out with a sense that great responsibility lay on the shoulders of those tasked with "overseer" roles—such as the abbess who directed life in the abbey or convent. If one examines, for instance, the best-known "handbook" on monastic life—Benedict's Rule—you find that the responsibility of the "overseer" is far greater than any individual person within the monastic order. They were, after all, accountable to no less than the persons of God for how they handled their responsibilities. This is surely a vivid contrast with today's frequently unaccountable, unreliable and even regulation-averse surveillance, whether in government departments or the activities of mega-corporations such as Amazon and Uber.
Of course, early medieval surveillance underwent major changes as social, cultural, political and economic changes occurred—processes of secularization, state formation, rationalization and the development of science-and-technology all played a part. But these occurred over centuries, not at the breakneck speed of contemporary socio-technical change. As a scholar, working from the early 1980s on surveillance issues, I have personally experienced the feeling of acceleration of the rate at which novel technologies and systems have emerged. Examining the social, cultural, political and economic dimensions of surveillance, it is clear to me that processes of elision of "phases" and of increasing generational misunderstandings and incompatibilities frequently occur.
However, by drawing on the work of medieval historians, I aim to explore what actually happened in surveillance terms, avant la lettre. Medieval surveillance certainly offers a different perspective! Examining the "eye of God" requires a radical rethink of surveillance and its discontents, which also means questioning our common assumptions about technological "progress" and its political-economic priority around the world. I don't know where the solutions lie, but studying surveillance over the longue durée rather than only during the modern period, when it was given its name, offers some refreshing insights. Medieval surveillance, avant la lettre, indicates some fascinating and significant origins of modern surveillance, starting, in my approach, with the "eye of God." This idea dominated surveillance practices from about the C5th to the C12th and even after it was modified through early forms of secularization, retained some of its original features. By the C18th rationalism was becoming dominant and Jeremy Bentham's famous "panopticon"—all-seeing—prison plan retained the idea of God's eye, only without the God reference. This was a crucial moment in the secularization of "God's eye."
Concluding Challenge
AI and surveillance are vitally joined; the one depends on the other. AI thrives on data collected—extracted—through surveillance. Surveillance is enhanced through new AI techniques. Human beings are not overall beneficiaries of this collusion. Machine "Learning" and increasingly "autonomous" machines—such as self-driving vehicles—indicate how machine-centred thinking operates, with its desired take-over of formerly human tasks—and the rewards in the hands of monopoly corporations and increasingly autocratic government.
This seems exactly the wrong way round. Human-centred thinking should surely determine what machines are best-suited for enhancing human lives, and what place surveillance has within that equation. And human-centred action is called for in seeking democratically-oriented alternatives to current practices, especially in AI surveillance. One way to discover such possibilities is to examine how surveillance worked in pre-modern, medieval times, when its features were for human benefit. It was caring, justice-oriented, relational and reliable—four features now largely missing from today's surveillance, AI-supported or not.