What this threat is
Traditional surveillance required human attention. Watching a camera feed meant a person had to sit in front of a monitor. Reading communications meant analysts had to read them. Following a suspect meant deploying physical personnel. Those resource constraints weren't incidental; they were the practical limit that kept surveillance from becoming total even in states that would have welcomed the capability. AI eliminates the human attention bottleneck. Facial recognition systems can process thousands of camera feeds simultaneously, matching faces against databases in real time without any human involvement. The cost per monitored individual collapses to near zero at scale.
The technologies involved aren't limited to cameras and facial recognition. Gait analysis systems can identify individuals from how they walk, even when their faces aren't visible. Emotion recognition systems claim to infer emotional states from facial expressions, voice patterns, or physiological signals, and are being deployed in employment screening, border crossings, and educational settings. Social credit systems aggregate behavioral data across multiple domains, from financial transactions to online activity to physical movement, to generate composite scores that determine access to services, jobs, or travel. Each of these technologies is already deployed somewhere in the world at operational scale.
What makes AI surveillance qualitatively different from earlier forms isn't just quantity but the nature of what becomes knowable. When surveillance was limited by human attention, there was a practical ceiling on how much anyone could know about a given individual. AI systems that aggregate data across sources (movement patterns, purchase history, communication metadata, social networks, search behavior, physical appearance changes) can construct detailed behavioral profiles of individuals that reveal things those individuals don't know about themselves. They can identify political associations, health conditions, romantic relationships, financial stress, and psychological vulnerabilities through inference rather than direct observation.
The chilling effect is one of the most significant and least visible consequences. When people know or believe they're being watched, they change their behavior. They attend fewer political meetings, write less provocatively, associate less freely. Research on behavioral change under surveillance consistently shows that awareness of monitoring causes people to self-censor and conform, even when they're doing nothing wrong and even in societies that nominally protect civil liberties. A society where comprehensive surveillance is technically possible, even if it isn't universally deployed, is one where the population behaves as if it might be. That constraint on free behavior has costs that are diffuse and hard to measure but real and cumulative.
Why it matters
Privacy isn't a luxury preference. It's a structural requirement for the kind of free behavior that democratic society depends on. People form unpopular opinions, explore heterodox ideas, organize around minority interests, and engage in the normal messiness of political life only when they believe those activities aren't being permanently recorded and potentially used against them. Mass surveillance changes the structural conditions under which political life happens. It doesn't have to be actively weaponized to have an effect; the possibility of it being weaponized is enough to suppress behavior across a population.
The authoritarian application of AI surveillance is already well-documented. There are places in the world where comprehensive AI-driven monitoring of ethnic and religious minorities, political dissidents, and civil society organizations is operational and has been for years. The scale and sophistication of these systems continues to grow. But the risk isn't limited to states that are already authoritarian. Democratic societies are deploying many of the same technologies for purposes that seem reasonable in isolation (crime prevention, border security, fraud detection) and creating surveillance infrastructure that future governments, or current governments under different circumstances, could repurpose. The technology doesn't know the difference between a state that intends to protect rights and one that intends to suppress them.
Corporate surveillance presents a different but overlapping concern. Advertising technology, loyalty programs, smart devices, and mobile applications have created comprehensive behavioral monitoring systems operated by private companies. These systems profile individuals in ways they'd likely object to if they understood the detail, selling access to those profiles for purposes ranging from targeted advertising to political micro-targeting to insurance pricing. The data these systems collect is often available to governments through legal processes, subpoenas, or informal cooperation, meaning the practical distinction between corporate and state surveillance is less clear than it appears. A private company that knows someone's daily movements, health queries, and political interests has created a tool of control whether or not it intends to use it that way.
Where things stand today
AI surveillance systems are operational across a wide range of countries at varying scales. Mass facial recognition in public spaces, AI-driven social media monitoring for dissent, predictive policing algorithms, and automated border screening using behavioral analysis are all deployed somewhere in the world right now. The commercial market for these technologies is global, and export controls are limited. Several major commercial surveillance technology providers have sold systems to governments with poor human rights records with few restrictions, exporting not just the technology but the governance model that comes with it.
The EU AI Act is the most significant regulatory response to date. It prohibits real-time remote biometric identification in public spaces in most contexts, with narrow law-enforcement exceptions requiring judicial authorization. It prohibits emotion recognition in employment and educational settings. It bans biometric categorization systems that infer sensitive characteristics like political opinion, religious belief, or sexual orientation. These are meaningful restrictions, among the strongest in any major jurisdiction's AI regulation. But they apply only within the EU, and enforcement will depend on regulatory capacity and political will that haven't yet been tested at scale.
Where the regulation is still weak is in the areas of predictive policing, behavioral profiling, and corporate surveillance that doesn't obviously fall into the prohibited categories. The Act creates oversight obligations for certain high-risk applications but doesn't address the fundamental data collection infrastructure that makes surveillance possible. The export of AI surveillance technology from EU-based companies to third countries also remains largely unaddressed. International governance of surveillance AI is at an early stage, with general human rights norms providing some framework but no enforcement mechanism adequate to the scale of what's being deployed.
How Better Societies helps
Compliance: The EU AI Act's prohibitions on remote biometric identification, emotion recognition, and biometric categorization create real legal obligations for organizations operating in Europe. Our compliance programs help organizations understand which of their AI systems fall into prohibited or high-risk categories, what the specific obligations are, and how to build the documentation, risk assessment, and human oversight processes that the regulation requires. We work with organizations across sectors, including law enforcement adjacent contexts, HR technology, and retail analytics.
Summit: Surveillance governance requires norms that cross national borders, because surveillance technology and its effects don't respect them. The Better Societies Summit convenes policymakers, technologists, civil society organizations, and researchers to develop shared frameworks for what constitutes acceptable AI surveillance and how international standards can be built and enforced.
Accelerator: Privacy-preserving technology, oversight tools that give individuals visibility into how they're being monitored, and systems that enable accountability for surveillance deployments are all areas where innovation can shift the power dynamic. If you're building in this space, the Better Societies Accelerator supports founders who are working to give individuals more control over their digital environments and to make surveillance systems more accountable to the people they affect.