Facial Recognition — The Control-Grid Component
Facial recognition is the layer that makes anonymity in public optional. Once a camera network is wired to an identity database, the question “who is in this crowd” becomes a query, and presence at any place and time becomes a record. Every jurisdiction frames it differently — public safety, terror prevention, finding missing persons, catching fare-dodgers — and every jurisdiction is building it. Scored in the Convergence Index.
China
China runs it nationwide. The SkyNet and Sharp Eyes camera programs put the country at the top of every per-capita surveillance index (Comparitech), wired to the mandatory national identity layer so a face resolves to a name, a household, and a behavioral record by default. There is no warrant requirement and no independent oversight body to appeal to. Capability, scale, and unconstraint all max out — the reference implementation everyone else is measured against.
European Union
The EU is the case study in “banned with exceptions.” The AI Act prohibits real-time remote biometric identification in public — and then carves out exceptions for serious-crime search, missing persons, and imminent threats (AI Act, Regulation (EU) 2024/1689). Member states are already expanding deployment inside those carve-outs: France passed Law no. 2023-380 authorizing algorithmic video surveillance for the 2024 Olympics, the first such law in the EU (Légifrance). The ban is real, which is why the EU does not score at the top — but the exceptions are where the capability lives, and they only ever widen.
India
India is deploying facial recognition with essentially no governing framework — the gap between capability and limit is the widest of the four. The Internet Freedom Foundation’s Project Panoptic tracks dozens of government FRT systems rolled out across policing, airports, and welfare delivery with no enabling statute and no data-protection floor (Project Panoptic). The capability is broad and growing; the absence of any legal limit is the story, which is why India scores high on unconstraint despite lower comprehensiveness than China.
United States & United Kingdom
No federal ban, and the capability is largely privatized into the identity layer. Clearview AI scraped billions of public images into a face-search tool sold to US law-enforcement agencies — fined €30.5M by the Dutch DPA, which did not stop the US business (Reuters). The UK runs it through the state directly: the Metropolitan Police operate Live Facial Recognition under their own published policy, scanning crowds against watchlists in real time (Met Police). Fragmented and patchwork at the federal level, but operational on the ground — and the watchlist that defines a “match” is set by the deploying agency.
The counter-argument
The defenders’ case is concrete: facial recognition has identified trafficking victims, exonerated the wrongly accused, and found missing children. The problem is not that it never works — it is that it works probabilistically against everyone, all the time, with documented racial error gradients (the basis for the ProPublica and NIST findings that drove several US city bans) and no consent step. A tool that finds a missing child by scanning every face in the station has, in that same scan, recorded every other face too. The component does not distinguish the searches.
Part of the Convergence Index component set. Scored in the interactive index; full cross-country comparison in the convergence table.