UK smart lamp-posts with facial recognition spark surveillance fears

Conflow Power Group, based in Warwickshire, has deployed what it calls iLamps—solar-powered lamp-posts equipped with cameras for number-plate and facial recognition—in a hospital car park in the region. These represent the first such installation in the UK. CPG claims the lamp-posts can function not only as surveillance devices but also as "revenue-generating distributed AI datacentres," though the exact meaning and operational status of this claim remain unclear. The company has announced that it has signed an agreement with the government of Katsina state, Nigeria, to supply 50,000 iLamps with "additional surveillance potential for both vehicles and faces." Edward Fitzpatrick, CPG's chairman, told BBC World Service's Tech Life programme that the lamp-posts could ultimately track missing persons or fight crime, provided this capability were developed in partnership "with the relevant authority." CPG has stated it has received "extensive worldwide media coverage" but does not take further questions, directing journalists to its online material. The Warwickshire hospital and local authority have declined to provide additional details.

Privacy advocates and technical experts are deeply skeptical of both the feasibility and wisdom of the deployment. Jasleen Chaggar, senior legal and policy officer at Big Brother Watch, distinguishes facial recognition from traditional CCTV: whereas CCTV allows passive recording that is reviewed after the fact, facial recognition performs real-time identity checks and is "much more like a fingerprint or DNA." She warns that once infrastructure is in place, new surveillance functions—gait analysis, suspicious-behavior detection, aggression monitoring—can be added without a separate public decision: "a lot of these technologies and infrastructures have capacity to add function later." She cites Hammersmith and Fulham's planned CCTV upgrades as an example, where facial recognition and behavioral monitoring could be embedded and activated with minimal transparency.

Prof Rabih Bashroush, director of the Centre for Artificial Intelligence and Data Innovation at the University of East London, is blunt about the technical claims: "AI datacentres? There's absolutely no chance this will fly: the technology doesn't work, there's not enough power, the economics won't work." Solar power sufficient to run an AI datacentre does not exist in Britain—the energy generated by lamp-posts is "negligible." Small, non-critical applications like parking-spot occupancy detection might work, but claims about facial recognition for public safety are "just hype." Bashroush also notes a structural problem in the tech industry: many AI and engineering experts now work for companies with financial stakes in these systems and cannot publicly contradict vendor claims, leaving local authorities without independent guidance and the public unaware that deployments are happening.

Concerns about facial recognition accuracy compound these doubts. Chaggar notes that much "emotion-detection" and "suspicious-behavior" software is "pseudo-scientific," with little evidence of accuracy. Disabled persons who move differently could be flagged for unwarranted intervention. In retail, where facial recognition is already widely used to prevent shoplifting, misidentification can result in unexplained blacklisting, creating what Chaggar calls a "Kafkaesque" experience. Police have also misidentified people using facial recognition.

Data ownership presents another unresolved risk. Bashroush recounts a case from his work with the European Commission's Internet of Things initiative: a supplier collected garbage data from homes in the Netherlands and optimized truck routes based on residents' habits. When the local council tried to reopen bidding a year later for a better price, the company claimed ownership of the data, citing poorly constructed contracts. The authority lost access to the data it had generated, setting no precedent for how to handle such disputes going forward.

Politically, the timing is fraught. Home Secretary Shabana Mahmood has publicly endorsed a "panopticon state"—one that can watch citizens at all times—and framed this in terms of security, telling BBC 5 Live that "you can't enjoy any of your liberties in any country if you're not safe." This logic subordinates all civil liberties to state protection. The concern, voiced in the article, is that this infrastructure, once built and normalized, will persist even if future governments are less democratic or more authoritarian. Interviews with residents in Warwickshire reveal varied attitudes: most express comfort with CCTV as an established fact of modern life and skepticism that facial recognition represents a meaningful additional threat. However, anxiety about AI's broader impact on employment and society surfaced among some respondents, such as Tejal, 41, a radiologist, who expressed deep concerns about AI's capability in her own field and what she feared it meant for her children's futures.

The iLamps represent a convergence of two long-standing tensions in surveillance policy: the normalization of CCTV as a taken-for-granted part of urban life, and the step-change in capability that facial recognition introduces. As Jasleen Chaggar of Big Brother Watch emphasizes, facial recognition is fundamentally different from CCTV—it performs real-time identity checks rather than passive recording reviewed after the fact, making it closer to a biometric fingerprint. The danger, she argues, is that once the physical infrastructure exists (as Hammersmith and Fulham's updated CCTV demonstrates), new functions can be "turned on overnight" without requiring a separate public decision or consent. This matters because the political context is permissive: the home secretary has publicly endorsed a "panopticon state," and experts worry that future governments, potentially even more surveillance-friendly, will inherit ready-made infrastructure designed to watch citizens continuously.

The article also exposes a gap between engineering feasibility and vendor marketing claims. Prof Bashroush's critique—that AI datacentre functions rely on insufficient solar power and don't work in Britain's climate—suggests that Conflow Power Group may be packaging a modest CCTV upgrade with tech buzzwords to attract investment or government contracts. This pattern, he warns, reflects a wider problem: many tech experts now work in industry and cannot publicly contradict vendor claims, while local authorities lack the technical knowledge to evaluate them and the broader public remains unaware. The data-ownership issue, illustrated by the Netherlands garbage-collection example, adds another layer of risk: once a company controls the data stream, local authorities may find themselves locked in, unable to renegotiate terms or switch vendors.