The Invisible AI Map
In January 2020, Detroit police arrived at Robert Williams’s home in Farmington Hills, Michigan, and arrested him in front of his wife and their young daughters. He was accused of stealing watches from a Shinola store in downtown Detroit two years earlier. He had not done it. A facial recognition search had wrongly pointed police to him (Michigan Public, 2024).
His case ended in a settlement, announced in June 2024, that put new layers of safeguards on when and how Detroit officers can use facial recognition. It changed how lineups are conducted, required officer training, required police and prosecutors to tell defence lawyers when facial recognition was used, and gave a federal court four years of oversight (Michigan Public, 2024). The department now says a match “can only be used as a tip to further an investigation.”
Mr Williams never agreed to have his face searched, and did not know it had been until officers were at his door. That is the heart of what makes facial recognition in public spaces different from almost any other technology you meet in daily life: it acts on you without asking, and often without telling you.
So here is my proposition: the real danger is not the camera. It is acting on an unchecked match about someone who was never told. The shop or police force gets speed. When the match is wrong, you pay.
How a face becomes a suspect
Facial recognition systems take an image of a face, turn it into a mathematical template and compare it with others. In public spaces it is used in two broad ways. Shops, stadiums and venues may compare everyone who walks in against a watchlist of people they want to keep out. Police may compare faces in a crowd, or in a still from a security camera, against databases of people they are looking for.
When it goes wrong, the consequences land on a real person. Shops have made mistakes too: in 2023 the FTC banned the pharmacy chain Rite Aid from using facial recognition for surveillance for five years, saying its system had falsely flagged customers as shoplifters, particularly women and people of colour (FTC, 2023).
Key idea: Verification versus identification
Verification asks a narrow question: is this the person they claim to be? Unlocking your phone with your face is verification, and you chose it. Identification asks a much bigger one: who, out of everyone in a database, is this? Scanning a crowd is identification, and every face is searched whether or not its owner agreed. Identification makes more comparisons, so even a small error rate produces false matches, and each false match is a real person who may be stopped, questioned or turned away.
The evidence room: what the law actually says
The rules depend heavily on where you are. A brief, non-exhaustive guide:
European Union. The EU AI Act bans police from using real-time facial recognition in publicly accessible spaces, except in narrowly defined cases such as searching for a victim of abduction or trafficking, preventing an imminent threat to life or a terrorist attack, or locating a suspect of a serious crime. It also bans building face databases by untargeted scraping of images from the internet or CCTV. These prohibitions have applied since 2 February 2025 (Regulation (EU) 2024/1689, arts. 5 and 113). Your data protection rights under the GDPR also apply.
United Kingdom. There is no single law written specifically for facial recognition. Data protection law applies, and the Information Commissioner’s Office (ICO) handles complaints. You can ask an organisation to delete data it holds about you, and it normally has one calendar month to respond (ICO, n.d.-b). The Home Office consulted from December 2025 to February 2026 on a new legal framework for police use of facial recognition and is still analysing responses (Home Office, 2025).
United States. There is no federal facial recognition law, so protection is patchy. The FTC can act against unfair practices by businesses, and some police departments, like Detroit’s, now have written limits on how a match may be used. Some cities and states have their own rules. In New York City, for example, shops, restaurants and entertainment venues that collect biometric information must post a clear sign near every customer entrance, and they may not sell that information (NYC Local Law 3, 2021).
Australia. Biometric information is sensitive information under the Privacy Act. In 2026, the Administrative Review Tribunal affirmed aspects of the Privacy Commissioner’s finding against Bunnings over facial recognition in 62 stores, and the regulator now says “a precautionary approach” is required (OAIC, 2026). A separate case about Kmart is still before the tribunal.
The complication: even the strictest law says “sometimes”
Here is the best argument against me. Even the EU’s rule, the toughest here, does not ban police face-scanning outright. It allows it to search for an abducted or trafficked victim, or to stop an imminent attack. Lawmakers decided that, in a real emergency, a face match can be worth the risk. Detroit, too, kept the technology and changed the rules: a match is now only a tip.
That is a serious position. But notice that nobody in these documents argues a match should be trusted on its own. The fights are about when, who checks, and whether you are told.
The verdict
The record supports the proposition. Robert Williams was not harmed by a camera but by a search result treated as an answer. The FTC said Rite Aid’s system falsely flagged customers as shoplifters. Where rules have tightened, they target exactly that gap: telling people, checking the match, and keeping a court or regulator watching. An organisation that cannot say whether it scans faces, or what happens after an alert, has not earned your trust.
Five questions that put you back in the picture
- At a shop or venue: Do you use facial recognition here, and where is that written down?
- If you are stopped or turned away: Was I flagged by an automated system, and did a person check the match before acting on it?
- To any organisation: Is my image or face template held on a watchlist, how long will you keep it, and how do I get it removed?
- To your local council or police force: Is live facial recognition used in this area, who authorised it, and where are the results of its accuracy testing published?
- If you run a business considering it: What happens, step by step, after the system raises an alert, and how will you know when it is wrong?
Where to complain
- UK: raise it with the organisation first, then the ICO (0303 123 1113).
- EU: your national data protection authority.
- US: the FTC at ReportFraud.ftc.gov, and your state attorney general.
- Australia: the organisation first, then the OAIC.
What we still don’t know
This is a general guide, not legal advice, and it does not cover every country or every local law. We describe the Detroit settlement as reported by Michigan Public; we have not reviewed the agreement itself, and it applies only to Detroit police. The FTC’s findings mentioned above were allegations settled by agreement, not a court verdict. We do not have reliable public figures on how many shops or police forces in any country use facial recognition today, because in most places there is no register. Accuracy varies widely between systems, settings and lighting, so a figure from one test tells you little about the camera above a particular door. And the law is moving: the UK’s framework and the Australian Kmart case are both unresolved.
Which leaves the question we will keep chasing: with no register, how would you ever learn your face was searched, unless the police turn up at your door?
For organizations: go further. A facial recognition alert is a decision-influencing system, whether or not a person signs off at the end. Operational AI Readiness, FairByDesign’s free field manual, covers how to classify such systems honestly in Chapter 3, and how to make human review real rather than a rubber stamp in Chapter 5, Human Oversight That Actually Works. Read it free, or test one workflow with the free readiness assessment.
Stay informed. The free Mercury Brief brings you plain-language guides like this, with the record behind them, in a short email. Get the free Mercury Brief.
FairByDesign is reader-funded. If this helped, get the free Mercury Brief or see ways to support. Support never buys editorial control.
References
Bakos, V. (2026). Operational AI readiness: A practitioner guide to governing AI that is already live. FairByDesign. https://fairbydesign.org/operational-ai-readiness/
Federal Trade Commission. (2023, December 19). Rite Aid banned from using AI facial recognition after FTC says retailer deployed technology without reasonable safeguards [Press release]. https://www.ftc.gov/news-events/news/press-releases/2023/12/rite-aid-banned-using-ai-facial-recognition-after-ftc-says-retailer-deployed-technology-without
Home Office. (2025, December 4). Legal framework for using facial recognition in law enforcement [Consultation]. GOV.UK. https://www.gov.uk/government/consultations/legal-framework-for-using-facial-recognition-in-law-enforcement
Information Commissioner’s Office. (n.d.-a). Data protection complaints. Retrieved October 15, 2026, from https://ico.org.uk/make-a-complaint/data-protection-complaints/data-protection-complaints/
Information Commissioner’s Office. (n.d.-b). Your right to get your data deleted. Retrieved October 15, 2026, from https://ico.org.uk/for-the-public/your-right-to-get-your-data-deleted/
Michigan Public. (2024, June 28). “It didn’t make sense at all”: Wrongful facial recognition arrest in Detroit leads to landmark settlement. https://www.michiganpublic.org/criminal-justice-legal-system/2024-06-28/it-didnt-make-sense-at-all-wrongful-facial-recognition-arrest-leads-to-landmark-settlement
New York City Local Law No. 3 of 2021, Biometric identifier information (N.Y.C. Admin. Code §§ 22-1201 to 22-1205). https://intro.nyc/local-laws/2021-3
Office of the Australian Information Commissioner. (2026, July 29). Privacy Commissioner publishes updated guidance on facial recognition in retail spaces [Media release]. https://www.oaic.gov.au/news/media-centre/privacy-commissioner-publishes-updated-guidance-on-facial-recognition-in-retail-spaces
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). (2024). Official Journal of the European Union, L, 2024/1689. http://data.europa.eu/eli/reg/2024/1689/oj



Leave a Reply
You must be logged in to post a comment.