Article

August 12, 2026

AI in the Public Eye

# Ai

# Artificial intelligence

# Cities

# Compliance

# Computer Vision

# Government

# Hafnia

# Responsible ai

# Responsible technology

# Smart city

# Data

# Mobility

# Traffic

# Transport

# Trust

Exploring the future of computer vision and responsible AI

Roland Harwood

Overview

Computer vision is all around us. It helps to manage a crowd at a festival, tracks a flood through a city, and reroutes traffic around an incident before congestion builds. Done well, it can save lives and time at a scale no human team could match, making our world safer, smarter and more sustainable.

None of that requires a single camera to identify you or infringe on your right to privacy. Yet the gap between genuine value and real risk underpins the current and future uses of AI and computer vision in the public realm and requires a more honest debate about the challenges and opportunities.

This article underpins a new community and series of Tech Talks called AI in the Public Eye, which are for anyone who cares about where computer vision and AI are heading in public life.

Trust is the real bottleneck

In the UK, as well as in several other countries, we police primarily by consent. It is a simple idea with a big consequence. The state's authority to act depends on the public agreeing, in general terms, that the action is legitimate.

91% see use of AI in policing as beneficial, yet 72% still say regulation would increase their comfort with AI.

We believe that AI needs similar foundations. Not permission from a regulator alone, but consent from the people whose faces, streets and daily lives are the raw material. Without that consent, deployment does not stick. With it, almost anything becomes possible.

Public trust is conditional, not fixed. In 2019, the Ada Lovelace Institute found 49% of Britons supported facial recognition in everyday policing, assuming safeguards, while 67% opposed its use in schools. By 2025, a follow-up survey with the Alan Turing Institute found 91% now see policing use as beneficial, yet 72% still say regulation would increase their comfort with AI. Support has grown, but it has grown alongside demand for guardrails, not instead of it. (Sources: Ada Lovelace Institute, "Beyond Face Value," 2019 ; Ada Lovelace Institute and Alan Turing Institute, "How Do People Feel About AI?," 2025 .)

The backlash has already started

Without consent, Orwell's Big Brother has migrated from fiction to a live product decision: whether a camera on a street corner should be able to identify the person walking past it. Are we blindly building god-like technology without understanding what is coming next?

Jeremy Bentham designed his 18th century Panopticon prison around one insight: people change their behaviour once they believe they might be watched whether anyone is. That insight still runs under the debate about computer vision today.

Therefore, people are pushing back in a variety of ways. One novel method is via adversarial clothing such as Cap_able , an Italian design and research project that creates textiles incorporating patterns designed to interfere with certain computer vision and object detection systems, potentially causing them to misclassify or fail to detect the wearer as a person, and instead identify a giraffe or a zebra instead. It sounds like a novelty. It is really a wearable form of protest, aimed at questioning the increasingly pervasive role of machine vision in public space, and priced at several hundred euros, aimed squarely at a market that has grown from under four billion dollars in 2024 towards an expected twelve billion by 2030.

Shoshana Zuboff gave this backlash a name: surveillance capitalism. Her argument, in short, is that our behaviour has become raw material for prediction products, extracted largely without our consent and traded in markets we never see. We agree to terms and conditions we do not read, in exchange for services that feel free. They are not. If you are not paying, you are the product.

The reality is municipalities (and companies) do pay to install computer vision to capture events based upon a social contract where people pay taxes (or other charges) in exchange for security and other services. That trade-off is precisely why consent matters more than convenience. Once people feel the asymmetry, the backlash follows, whether through fashion, legislation or refusal to engage.

That contract depends on honesty holding up on both sides. In July 2026, the ACLU documented how Flock Safety, a US computer vision company whose automatic number plate recognition (ANPR) cameras are used by police departments nationwide, had repeatedly misled city councils about its own system's capabilities, including denying it could track a vehicle's movements over time until forced to admit otherwise. One Wisconsin city, Oshkosh, approved a Flock contract, discovered the deception the next morning, and cancelled it within a day, the fastest reversal on record. Consent revokes fast once trust breaks, whether the computer vision in question reads a face or a number plate.

The utility case is real too

None of this means the technology is without merit. Transport for London and British Transport Police have just started trialling live facial recognition at London Underground stations , scanning for people wanted for sexual offences, robbery and knife crime. Non-matches are deleted immediately, and TfL has built the trial around informed consent, clearly signposted zones with alternative routes for anyone who would rather avoid the cameras. But the same reporting notes that roughly 530,000 face scans across UK railway stations since February have produced zero matches, zero arrests and one false identification, and Big Brother Watch has called it a "dystopian expansion" that treats the public like suspects. The debate about where the line sits, and whether the technology even works as well as claimed, is unresolved.

There is a better way to get the benefits without the costs. brighter AI , now part of Milestone and a contributor to Project Hafnia, has built Deep Natural Anonymization, a technique that detects faces and number plates in video and replaces them with synthetic equivalents. The result preserves everything useful about the footage for analytics and safety while removing the personal data that makes it a privacy risk. It is GDPR compliant by design rather than by afterthought, and it has already anonymised more than twenty billion images for clients in transport, automotive and smart cities.

This is what we are doing with Hafnia . We make real-world video data available to develop AI safely, compliantly and responsibly, using anonymisation as the default rather than the exception. This data is made available for AI developers to train and fine tune their models and underpins vision language models (VLMs) that produce accurate text-based summaries of what is happening in a video stream almost instantly. Utility and privacy are not opposites. They can be engineered to coexist.

Peta Bencana in Jakarta shows what this looks like when it works well. It combines official flood and disaster data with crowdsourced reports sent by residents over social media and messaging apps, verified by government agencies and turned into a live public map. It has engaged more than 200 million users and trained tens of thousands of people in disaster risk reduction. That is collective intelligence in its purest form: AI plus humans in the loop, neither one replacing the other. Computer vision has the same potential for traffic management, logistics and public safety, if we build it the same way.

What Living Labs are teaching us

At the Paris 2024 Olympics, French authorities enabled a time-limited, city-wide living lab, allowing AI-powered video surveillance from companies including our partner Wintics to flag events such as crowd surges, abandoned bags and weapons, under a law that explicitly excluded facial recognition. Officials called the trial a success and extended it, as I wrote at the time .

"Over 80% of citizens supported the use of AI in public spaces once they understood how it worked." Matthias Houllier

At the same time, digital rights groups including La Quadrature du Net challenged the system in court, arguing that continuous behavioural analysis is a form of biometric processing even without facial recognition, and that concerns about mission creep were far from resolved. So the lesson is not that the public simply accepted this once it was explained. It is that a narrow, time-boxed, legally constrained deployment can get further than an open-ended one, and that the argument about where the line sits does not end just because the cameras were switched off.

That is exactly what a living lab is for: a bounded, real-world space to learn what works and what doesn't before scaling up. We have been working with the Digital Outdoor Living Lab in the wider Copenhagen area, and the Houston Living Lab in Texas, to test real-world use cases and smart-city solutions in a live environment. We believe there is considerable opportunity to use these labs as test-beds for cross-city data sharing and innovation. If you would like to learn more, or be part of our growing community of pioneer places, get in touch and let's scope a partnership.

Building trust in practice

Two other examples are worth studying, precisely because they take alternative routes to the same goal. Amsterdam and Helsinki each publish a public AI register , a searchable list of every algorithm the city uses, what data trains it, and who is accountable if it goes wrong. Launched in 2020, they were the first cities in the world to do this. The bet is that transparency, offered upfront and by default, builds more trust than a good explanation offered after the fact.

Oakland, California, took a different route. Before any new surveillance technology can be bought or deployed by the city, residents and an elected Privacy Advisory Commission get a formal say first, not after. It is slower, and more adversarial, several technologies, including facial recognition, ended up banned rather than approved. But the underlying principle is the same one we started with: consent works better as a gate at the front door than an apology at the back.

And yet in our experience cities collect enormous amounts of video data and could make much better use of it. What they do collect rarely leaves the city boundary, even though the challenges are similar from one city to the next. Therefore, it's worth exploring how we can scale computer vision across cities and also asking how cities can invest the data they already have, and how residents can consent in a meaningful way.

Where we go from here?

The future is not fixed. It gets built by the decisions we make today and the questions we are willing to sit with.

We want to talk to and work with the pioneers building safe spaces to test these technologies, living labs where mistakes are cheap. We want to define what human rights look like in an age of AI, and who should be in the room when it gets written.

We are hopeful about our capacity to meet this moment. It asks us to keep our eyes, our minds and our hearts open at the same time, which is harder than it sounds.

These questions underpin a new series of Tech Talks called AI in the Public Eye. The first episode is live now and available to view here , is a conversation with Katja Sirazitdinova from Nvidia on scaling computer vision across cities.

Our next Tech Talk will be Marian Klee from brighter AI joins us to talk about protecting identities in public spaces, the themes running through this article: consent, anonymisation, and what GDPR-compliant computer vision looks like by design rather than by afterthought. Please do register now here if you are interested and available to join on the 14th September.

Lastly, if you are interested in any of the themes or questions explored in this article then please also join the Hafnia community here to shape the discussion and build the tools for the future of computer vision and responsible AI in the public eye.

This article is also reposted on LinkedIn here .

Responsible AI in the Public Eye