In short: The UK, the EU and the US are writing rules for facial recognition at three different speeds, and private-sector deployments have largely been left to govern themselves in the meantime. That gap is not permission. The three regimes already ask the same questions, and organisations that build to those questions now will not have to rebuild when the law catches up.
The referee’s arm is out: facial recognition and the price of playing on
by Richard Dutton, Lead, Data Compliance Practice, TL Agency & Consultancy

Police facial recognition is getting a statute. Private deployments will be drawn into the same Bill, but their rules are unwritten, Europe’s deadline has slipped and America has no federal law at all. Boards that read the gap as permission are misreading the game.
On 4 November 2026 the Home Office set out how it intends to govern police use of facial recognition. There will be a new law, carried in the Police Reform Bill; an independent regulator with the power to pause deployments; and a requirement that the technology be tested for bias by age, gender and race. For the police, the laws of the game are finally being written down.
For everyone else, the news attracted fewer headlines but matters more. The minister’s written statement to Parliament confirms that the same Bill will “also contain complementary measures” so that private-sector use of facial recognition to prevent and detect crime is “consistent with law enforcement use, where this is appropriate”, supported by a legal code for businesses. The Bill has not yet been introduced, no date has been given, and not a word of the code has been published. Across the Channel, the European Union has pushed the AI Act’s high-risk obligations — which catch remote biometric identification by private operators — back to 2 December 2027. In the United States there is still no federal law at all.
Read quickly, that looks like open field for the private sector until the new rules arrive. Read carefully, it is something else entirely, and anyone who has played rugby will recognise it.
Advantage is not a pardon
When a side infringes and the opposition can make something of it, a good referee does not blow straight away. He plays advantage. The arm goes out, play continues, and the offending side carries on as though nothing has happened. But the offence has been seen. If no advantage accrues, the referee brings play back to the mark of the original infringement and awards the penalty there.
That is precisely where private-sector facial recognition stands. Legislators and regulators in all three jurisdictions have their arms out. Play continues — in shops, stadiums, casinos and transport hubs — but the infringements are being noted, and for a simple reason: they are already infringements. The new statutes and codes will add detail, and in the UK the direction is now explicit: private use is to be held to the standard set for the police. But the law that matters is in force today. When a complaint, a claim or an investigation finally stops play, the penalty will come back to the mark: to the day the cameras were switched on without the answers a regulator expects.
Three referees, one rulebook
In the UK, a face processed to identify someone is special category biometric data under the UK GDPR, needing both a lawful basis and an Article 9 condition. The Court of Appeal’s judgment in Bridges (2020) found South Wales Police’s deployment unlawful largely because too much was left to individual discretion: who went on the watchlist, and where the cameras went. The Data (Use and Access) Act 2025 relaxed the rules on automated decision-making generally but kept the tightest restrictions where special category data is involved, alongside the right to human intervention and to contest a decision. And since 30 September 2026 the referee itself has changed: the Information Commission, a board rather than a single commissioner, has taken over from the ICO.
In the EU, the GDPR has treated biometric identification as special category data since 2018, and none of that has been deferred. The AI Act’s prohibitions have applied since February 2025, including real-time remote biometric identification by police in public spaces (with narrow exceptions) and the untargeted scraping of facial images to build databases. Private remote biometric identification is classed as high-risk, carrying obligations on risk management, data governance, logging, accuracy and human oversight; the Digital Omnibus moved those obligations to December 2027, but did not soften them.
The AI Act’s deadline moved. The GDPR’s never did. |
In the US, there is no federal statute, but the referees are hardly absent. Illinois’s Biometric Information Privacy Act requires written notice, written consent and a published retention schedule, and gives individuals a private right of action; it has produced some of the largest privacy settlements on record, Facebook’s $650 million among them. A 2024 amendment limited damages to one violation per person rather than one per scan, but left the duties intact, and Texas, Washington and others have biometric laws of their own. At federal level, the Federal Trade Commission banned Rite Aid in December 2023 from using facial recognition for surveillance for five years, after finding it had failed to test the system’s accuracy, train its staff or prevent false matches that fell disproportionately on women and people of colour.
What every referee is watching
The accents differ, but the questions are the same. Strip away the statutory language and every regime asks six things of anyone using a face to identify a person:
- Purpose and necessity. Why facial recognition, why here, and why not something less intrusive?
- The watchlist. Who goes on it, on what evidence, who decides, and how does anyone come off it?
- Accuracy and bias. Has the system been tested across a representative range of faces, in your lighting and your doorway, rather than in the vendor’s brochure? Bias is real; the only defensible position is to measure it.
- A human in the loop. Is anyone refused entry, refused service or accused on the strength of an algorithm alone, and can they challenge the decision?
- Are non-matches deleted at once, and alerts kept no longer than the purpose demands?
- Transparency and redress. Do people know the system is there, and is there a complaints route that actually works? In the UK, that has been a statutory duty since 19 June 2026.
Rite Aid fell on the third and fourth. South Wales Police fell on the second. Most Illinois claims turn on the sixth. None of those cases was ultimately about whether the camera worked.
Build once, to the strictest reading
For an organisation operating across borders, the temptation is to run three compliance programmes, each tuned to the referee of the day. That is expensive, and it is fragile: the moment one regime tightens, the others are out of step. The better course is to build once, to the strictest reading of the common rulebook, document every decision, and have the result tested against an external benchmark.
That is what standards are for. In the UK’s land-based gambling sector, where facial recognition is used to protect people who have asked to be excluded from gambling venues, the sector is now working on FRACT, the Facial Recognition Accreditation Compliance Technology standard: an open benchmark that turns “trust us” into evidence an operator can put in front of a regulator, a claimant or a board. The same discipline travels. Having spent four years guiding a retail deployer through a regulatory investigation, my experience is that the regulator was never chiefly interested in the camera. It wanted to know who was on the watchlist and why, how long images were kept, and what staff did when an alert fired. Those are governance questions, and they have governance answers.
3 questions for the next board meeting
Any board with facial recognition in its estate, or on its roadmap, should be able to answer three questions without notice:
- Could we explain to someone wrongly flagged exactly why the system flagged them, and what we did next?
- Do we know our system’s error rates in our own conditions, broken down by demographic group?
- If the law in our strictest market applied everywhere tomorrow, what would we have to switch off?
If the answers are uncomfortable, the time to find out is while play is still going on.
Advantage over
Rugby referees call “advantage over” when the non-offending side has gained enough that the original offence no longer matters. That call is not coming for the private use of facial recognition. The public, the courts and the regulators have seen too much of it to let the infringement go. The organisations that treat this period as free play will find the penalty brought back to the mark of their earliest mistake. Those that use it to build, and to prove, a defensible deployment will find that when the whistle finally goes, it is not blown against them.
Declaration of interest: Richard Dutton is Managing Director of ELIAS Partnership, which has commercial interests in facial recognition technology and in the FRACT standard referred to in this article.
Richard Dutton leads the Data Compliance Practice at TL Agency & Consultancy. He is Managing Director of ELIAS Partnership, which he co-founded with Dean Armstrong KC, and advises on facial recognition, biometric and special category data across gambling, retail and town-centre partnerships. A founding member of the Loyalty Academy’s Board of Regents (2015–24), he is a Certified Loyalty Marketing Professional.
Sources
- Written ministerial statement HCWS449, Sarah Jones MP, Minister of State for Policing and Crime, 4 November 2026:
- Home Office press release, “Stronger laws for police using facial recognition technology“, 4 November 2026:
- Government response to the consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies (October 2026), published on gov.uk.
- R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058.
- Information Commission succeeds the ICO, 30 September 2026: gov.uk announcement.
- Data (Use and Access) Act 2025: Articles 22A–22D UK GDPR (automated decisions); s.164A DPA 2018 complaints duty, in force 19 June 2026.
- EU AI Act (Regulation (EU) 2024/1689): Article 5 prohibitions, applicable from 2 February 2025; Annex III point 1 (biometrics).
- Digital Omnibus on AI, Regulation (EU) 2026/1744, in force 27 July 2026; Annex III obligations from 2 December 2027 (Hunton Andrews Kurth summary).
- Illinois Biometric Information Privacy Act, 740 ILCS 14; amendment SB 2979 (August 2024); In re Facebook Biometric Information Privacy Litigation ($650m settlement, 2021).
- Federal Trade Commission v Rite Aid, order announced 19 December 2023.
Researched and drafted with AI assistance, edited and fact-checked by the author. Illustration: AI-generated.



