Comprehensive surveillance was always limited by the cost of human attention. There were never enough watchers for all the watched. AI eliminates this bottleneck entirely. A single system can now monitor millions of faces, conversations, transactions, and movements simultaneously, at trivial marginal cost. The shift is qualitative: from selective monitoring, which targets known suspects, to ambient monitoring, which watches everyone and lets algorithms flag anomalies. Over one billion surveillance cameras are now installed worldwide, more than half of them in China, where an estimated 600 million cameras are integrated with AI facial recognition across the Skynet and Sharp Eyes systems (IHS Markit via CNBC; Video Experts Group). In the United States, the ratio is comparable: one installed camera for every 4.6 people, compared to one per 4.1 in China (CNBC/IHS Markit). The American figure includes millions of private doorbell cameras, which might seem categorically different from state surveillance infrastructure. It is not. In February 2026, Ring’s Super Bowl ad for its “Search Party” feature, which uses AI to scan outdoor cameras across entire neighborhoods to find lost dogs, triggered immediate backlash. Critics described it as a Trojan horse. The system required to identify and track a moving subject across multiple private cameras already exists; expanding what, or who, it tracks is a software choice, not a hardware upgrade (Inc.; GeekWire). Ring had simultaneously partnered with Flock Safety, a company that provides AI-powered license plate readers to law enforcement, before cancelling the integration under public pressure (Variety; Axios). The distinction between “private” and “state” surveillance dissolves when private infrastructure can be networked, searched, and shared with a policy change. A study published in The Quarterly Journal of Economics found that fewer people protest when public safety agencies acquire AI surveillance software to complement their cameras. The mere presence of the systems suppresses unrest, whether or not they function as advertised (Bulletin of the Atomic Scientists; Lawfare). Surveillance no longer requires a police state’s budget. It requires a software license.
This is not only a Chinese phenomenon. The Carnegie Endowment’s AI Global Surveillance Index found that Chinese tech companies supply AI surveillance technology to at least 63 countries, and over 70% of Huawei’s “safe city” agreements involve nations rated “partly free” or “not free” by Freedom House (Carnegie Endowment; Bulletin of the Atomic Scientists). Companies based in liberal democracies, including Germany, France, Israel, Japan, the UK, and the United States, also actively sell sophisticated surveillance equipment to authoritarian regimes (Carnegie). The research on democratic backsliding is sobering: mature democracies did not experience erosion when importing surveillance AI. Weak democracies did, regardless of whether the technology originated from China or the United States (Bulletin of the Atomic Scientists). The tool itself creates the gravitational pull toward abuse. The supplier’s flag is secondary.
Democratic nations face their own version, less dramatic but equally pervasive. In the United States, 74% of employers now use online tracking tools, including real-time screen monitoring and web browsing logs, while 75% monitor physical workplaces through cameras, badge systems, and biometric access controls (ExpressVPN survey, Feb 2025). The EU’s 2024 European Working Conditions Survey found that 42.3% of EU workers are affected by algorithmic management (European Parliament). The data broker industry generated roughly $280 billion in revenue in 2024, packaging behavioral, location, financial, and biometric data from billions of people into purchasable profiles (Grand View Research; Maximize Market Research). American cities from Los Angeles to Chicago to Pasco County, Florida, have deployed predictive policing systems that integrate surveillance feeds, gunshot detection, and algorithmic risk scoring. In Pasco County, more than 1,000 residents, including minors, were subjected to repeated police visits based on algorithmic predictions, a practice later found to have violated their constitutional rights (The Conversation; Brennan Center). The panopticon in democratic societies does not arrive through decree. It assembles itself through thousands of independent purchasing decisions, each individually defensible, that collectively produce comprehensive surveillance.
Even optimistic governance barely breaks even on this shade. The track record demands honesty. “Privacy by design” has been a stated principle since Ann Cavoukian coined the term in 1995. Three decades later, a 2022 European Commission study found that 97% of the most popular websites and apps used by EU consumers still deployed at least one dark pattern violating its principles (European Commission; Freshfields). The world’s most comprehensive privacy regulation moved the needle from near-universal violation to merely widespread violation. Security technologist Bruce Schneier has compared this moment to the early Industrial Revolution: pollution was also easier to deploy than to regulate, and we did not solve it with a single architectural principle (Schneier, SecurityWeek). We solved it over decades through a composite of specific bans, liability frameworks, market incentives, cultural shift, and institutional enforcement. Surveillance governance is at the “dumping pollution in the river” stage. The EPA equivalent does not yet exist.
The most instructive case may be WhatsApp. After the Snowden revelations made mass surveillance a public concern, WhatsApp partnered with Open Whisper Systems in late 2014 to deploy the Signal protocol, giving two billion users end-to-end encryption that Meta itself cannot break (Wiley/Johansen et al., 2021). This is a company whose parent earns over $160 billion annually from targeted advertising, voluntarily encrypting its core product in a way that prevents it from reading the content. The reason was not regulation. It was competitive pressure from Signal and Telegram, amplified by public awareness. The proof came in January 2021, when WhatsApp announced data-sharing with Facebook and millions of users defected overnight. Signal added 4.6 million users in four days. WhatsApp reversed the decision (Appfigures). By May 2025, Meta was running its largest-ever marketing campaign, “Not Even WhatsApp,” built entirely around the promise that no one, including Meta, can see your messages (The Drum; Campaign). The feedback loop is real: awareness creates demand, demand creates competitive pressure, competitive pressure forces architectural adoption, and marketing reinforces the constraint by making it a brand promise that is costly to break. The constraint holds because the public is watching. The limits are equally real. Meta still collects extensive metadata: who messaged whom, when, how often, from where. It introduced ads to WhatsApp in mid-2025. The encryption protects message content. It does not protect the pattern of life surrounding the messages. Architectural constraints hold only where public awareness is specific enough to create competitive pressure. Message text got encrypted. Metadata did not. The camera on the doorbell got sold as a convenience. The surveillance network behind it received no equivalent scrutiny.
The deepest difficulty is structural: the entity best positioned to enforce surveillance constraints is the government, and the government is also the entity most tempted to use surveillance. The EU’s AI Act, which bans real-time remote biometric identification in public spaces for law enforcement with limited exceptions (HKFP/AFP), represents one early attempt to manage this tension. That tension does not resolve. It has to be managed, continuously, through the same separation-of-powers logic that governs every other concentration of state authority.
Key tension: The same technologies that enable state repression also enable personalized medicine, crime prevention, and disaster response. AI surveillance is simultaneously the most powerful tool for public safety ever created and the most powerful tool for political control.