A Geneva warning, a Geneva rulebook
On 23 June 2026, the UN Human Rights Council's 62nd session held an interactive dialogue on a report by Gina Romero, the Special Rapporteur on freedom of peaceful assembly and association, titled Unmasking the chilling effects of the digital surveillance ecosystem (A/HRC/62/45). Drawing on testimony from 152 activists, civil society representatives and journalists and written submissions covering 84 countries and territories, the report documents how live and retrospective facial recognition, alongside spyware and social media monitoring, is being used by law enforcement worldwide to identify, track and detain protesters — producing what Romero calls a systematic chilling effect on assembly rights, with civil society actors in "nearly every region of the world" now assuming they are watched.
The timing is pointed: the Council convenes in Geneva, in a country whose own federal police force operates under one of the more disciplined legal constructions of facial recognition anywhere. Fedpol's modernization of the national fingerprint database, AFIS2026, adds a facial-image comparison module funded by roughly CHF 25 million in Federal Council appropriations. Fedpol's own description of the system is explicit about its limits: matching draws exclusively on images already lawfully held in the AFIS database — not social media, ID photos or third-party sources — and, critically, "automatic real-time monitoring of persons (live scan) ... is excluded." Searches run only after a crime has occurred, against a fixed evidentiary record, with a human expert reviewing every match. Switzerland is not rejecting the technology; it is confining it to a narrow, retrospective, judicially accountable use case.
The case for the technology, stated fairly
Romero's report deserves to be met on its strongest terms rather than dismissed as reflexive tech skepticism. Law enforcement agencies have a legitimate interest in identifying suspects who commit violence or vandalism within large crowds, where witness identification is often impossible and video evidence is the only practical lead. Facial matching against a bounded database of existing criminal records — the Swiss model — can resolve cases that would otherwise go unsolved, and courts in most jurisdictions already accept CCTV and photographic evidence obtained at public gatherings. A blanket prohibition on any biometric matching, regardless of safeguards, would forfeit real investigative capability for a marginal privacy gain once robust use limitations are in place. The debate that matters is not "facial recognition: yes or no" but which of two very different deployment models a given country is actually running.
What "live" deployment looks like in practice
That distinction is exactly what recent reporting from Delhi illustrates. Delhi Police's AI-enabled camera network, whose latest phase Union Home Minister Amit Shah inaugurated in February 2026, brought 2,100 AI-enabled cameras online with a stated plan to add 10,000 more atop roughly 15,000 already-integrated older cameras. MediaNama reported that during the Cockroach Janata Party protests at Jantar Mantar in July 2026, journalists and bystanders documented vans equipped with cameras actively capturing and cross-referencing protesters' faces in real time — the precise "live scan" use case fedpol's own guidance rules out by design. The News Minute reported that India has no dedicated statute governing facial recognition technology, that Delhi Police did not conduct a privacy impact assessment before deploying it, and that a Public Interest Litigation is now before the Delhi High Court alleging protesters were continuously photographed and videographed for purposes unrelated to any specific offense. This is the surveillance architecture Romero's report is describing at the global level, playing out in real time in one of Asia's largest democracies.
Proportionality, not prohibition
Our editorial position remains that outright bans on biometric identification tools are the wrong lever — they forfeit genuine crime-solving capacity and are difficult to enforce once vendors and integrations exist regardless. But the Swiss and Indian examples, sitting side by side in the same news cycle, show what a workable regulatory floor actually requires: a specific statutory basis for the tool (not administrative discretion), an explicit prohibition on real-time or predictive use at assemblies, restriction of comparison databases to lawfully collected criminal records rather than open-ended social or ID data, mandatory human review of matches, and a prior, published privacy impact assessment before deployment — the precise step Delhi Police skipped. None of these conditions require banning the underlying technology; all of them are currently absent from India's framework and present, however imperfectly, in Switzerland's.
The UN Human Rights Council does not have enforcement power, and A/HRC/62/45 will not itself change a single procurement contract. What it does is put a comparative baseline on the table at the exact moment several Asian police forces are scaling live camera networks faster than their legislatures are writing rules for them. Switzerland's model is not perfect — critics reasonably ask whether "retrospective only" survives the same technical capability sitting on the shelf — but it is a legible standard other jurisdictions can be measured against. Regulators serious about keeping facial recognition as an investigative tool rather than a surveillance dragnet should be borrowing fedpol's constraints, not debating whether Geneva's human rights bureaucracy has jurisdiction to say anything about Jantar Mantar at all.