France's Parliament is debating whether to permanently install AI-enhanced cameras across public spaces — a significant expansion of the surveillance state that arrives with oddly thin empirical backing.
On May 12, 2026, Senator Pierre-Jean Verzelen (Horizons, Aisne) filed PPL 25-609, a proposition de loi titled Généraliser la vidéoprotection algorithmique — a bill to make algorithmic video surveillance (VSA) a permanent fixture of French law enforcement infrastructure. The bill proposes inserting a new Article L.251-1-1 into France's Internal Security Code (Code de la sécurité intérieure), authorizing AI systems to detect predetermined security events in real-time from camera feeds controlled by police, gendarmerie, transit authorities (SNCF, RATP), and fire services.
The Verzelen bill does not exist in isolation. It is the fourth of at least five legislative initiatives on VSA filed within a single twelve-month period — a fragmentation that civil liberties group La Quadrature du Net argues is deliberately structured to prevent comprehensive parliamentary deliberation on any single measure. The most consequential of these bills, the RIPOST law (Réponses Immédiates aux Phénomènes troublant l'Ordre public), passed the Senate on May 26, 2026, by 243 votes to 33, and arrived at the National Assembly on June 22, 2026. RIPOST extends algorithmic surveillance authorization through December 31, 2030 — covering the Alpine Winter Games — and expands its scope to include the interiors of buildings open to the public.
What the Technology Actually Does
Both the Verzelen bill and RIPOST explicitly prohibit biometric identification and facial recognition. The systems are designed to detect predetermined events — an abandoned package, a person lying on the ground, unusual crowd density, an apparent weapon — and generate an alert for a human operator to assess. Algorithmic outputs constitute "attention signals" only: they cannot independently support prosecutions or individual legal decisions, and activation requires prior prefectural authorization and CNIL-reviewed government decrees.
This distinction matters for EU AI Act compliance. Article 5(1)(h) of the Act, which entered into force in February 2025, prohibits "real-time remote biometric identification systems" in publicly accessible spaces for law enforcement purposes — with narrow exceptions for missing persons, imminent terrorist threats, and suspects in crimes carrying at least four years' imprisonment. Because French VSA systems detect events rather than identify persons biometrically, they do not fall squarely within this prohibition. They remain high-risk AI under Annex III of the Act when deployed for law enforcement, however, triggering requirements for fundamental rights impact assessments, independent auditing, and registration in an EU database — obligations whose fulfilment in France no court has yet been asked to scrutinize.
Steelmanning the Case for Expansion
The security rationale for expansion is not trivial. A human operator monitoring a live camera feed can sustain full attention for roughly twenty minutes before cognitive fatigue sets in — Verzelen's own legislative memorandum cites this physiological limit to argue that AI alerts are a force multiplier, not a replacement for human judgment. France already has approximately 90,000 cameras on its public roads; in many cases, the algorithmic processing hardware is already embedded in deployed units, awaiting activation. The argument is that this technology will be deployed regardless, and that a clear legal framework with explicit prohibitions on biometrics and mandatory oversight is preferable to the grey-zone deployments already underway.
France's February 2026 retail VSA law illustrates this logic. French startup Veesion had installed gesture-detection cameras in 2,000 to 3,000 stores before any specific law authorized it. Regularizing those deployments with safeguards — prefectural sign-off, biometrics ban, mandatory human review — was a proportionate response to demonstrated market reality, not a gratuitous surveillance expansion.
The Performance Gap No One Wants to Discuss
The steelman runs into a stubborn empirical problem. The Paris 2024 Olympics VSA deployment covered 185 cameras across Olympic sites. An independent post-event evaluation committee's January 2025 report found that four of the eight authorized use cases "did not function" as intended, and that the remaining four were "not particularly useful." Weapon detection triggered false positives when spectators opened umbrellas. Algorithms flagged homeless individuals as "fallen person" threats; park benches were misidentified as abandoned packages.
These findings were transmitted to Parliament. Yet Interior Minister Laurent Nuñez — who also authored RIPOST — publicly characterized the Olympic deployment as a success. La Quadrature du Net documented the gap: Nuñez "assured the national representation that these technologies had given complete satisfaction," a characterization the committee's actual findings do not support. A 2020 Court of Accounts report had already established no systemic correlation between camera density and crime reduction in France.
The CNIL's Measured Alarm and the LAPI Expansion
France's data protection authority, the CNIL, has consistently endorsed the limited experimental VSA framework while warning against its normalization. Its 2022 position statement called for "fixed red lines" — no biometrics, no automated legal decisions, no database cross-referencing. In its 2023 assessment of the Olympics law, CNIL described the introduction of VSA as "a turning point which will contribute to define the general role attributed to these technologies" — a careful warning that incremental normalization can preclude the deliberate societal choice a permanent architecture would represent.
RIPOST's parallel expansion of license plate reader (LAPI) data retention — from 15 days to one year, generating an estimated 700 million plate scans annually through the centralized STCL system — drew a separate CNIL warning about constructing a mass movement surveillance infrastructure through the back door of a security bill focused nominally on public order incidents.
Where France Ends Up
The Verzelen bill will not pass this session — PPL 25-609 has no scheduled debate date. But its filing marks a clear legislative intention: make VSA a permanent, end-date-free feature of French public space rather than a succession of temporary exceptions. RIPOST, now at the National Assembly, may deliver that outcome de facto through its 2030 extension and new indoor deployment authority before any formal vote on permanence occurs.
Five surveillance bills in twelve months, each incremental, none comprehensive. France may reach a permanent algorithmic surveillance architecture before any court or independent body has been formally asked to answer whether the evidence actually justifies it.