UK police admit their AI is biased, but promise to fight it

When a senior police official openly says that a technology the state plans to roll out at scale contains systemic bias, it deserves close attention. That is what happened in the UK, where Alex Murray, head of AI at the National Crime Agency (NCA), told the Guardian in an exclusive interview what is plain to see: AI used in law enforcement will be biased. He immediately added that efforts will be made to counter it.

The context of the statement matters as much as its content. The UK's Labour government has set a course for a dramatic expansion of AI use in policing in England and Wales. This is not a set of local experiments but a systemic transformation: police leadership is convinced that without artificial intelligence, law enforcement simply will not be able to keep up with the evolution of crime. To deliver on these ambitions, a dedicated police AI centre is being created with a budget of £115 million, roughly 13.5 billion rubles at the current exchange rate. That figure signals serious intent.

Murray rushed to reassure the public with a phrase that has already become a meme in the British media: "This is not RoboCop." According to him, the focus is above all on making complex investigations more efficient: processing large volumes of data, spotting patterns, speeding up analytical work. It sounds reasonable and even harmless. But the devil, as always, is in the details.

Bias in AI used for law enforcement is not an abstract, theoretical threat. Experience around the world has already produced plenty of troubling examples. In the US, predictive policing systems have repeatedly shown racial bias, directing disproportionately more resources to neighbourhoods where minorities live. Facial recognition systems have shown a markedly higher error rate when identifying people with darker skin. Recidivism risk-assessment algorithms have given higher scores to members of certain ethnic groups, all other things being equal. None of this is hypothetical; these are documented cases.

Murray's open acknowledgement of bias can be read two ways. On one hand, it is a show of intellectual honesty, rare among officials pushing costly technology projects. At this stage, the usual line is about the "neutrality of algorithms" and the "objectivity of data." On the other hand, the formula "we know there is a problem and we will work on it" is a classic expectation-management move that lets you deploy the technology now and deal with the problems later. A promise to "limit unfairness" is not the same as a promise to eliminate it.

For the wider context, it is important to understand that the UK is moving with a global trend. Law enforcement agencies around the world are expanding their use of AI, and the question is no longer whether police will use these technologies but on what terms. The European Union's AI Act sets strict limits on the use of real-time biometric identification systems. Post-Brexit, the UK is not bound by these rules and, by all appearances, is choosing a more liberal approach that emphasises self-regulation and internal oversight.

The architecture of oversight deserves separate attention. Creating a dedicated centre for £115 million is an attempt to centralise development and supervision, which in theory is better than individual police units adopting AI haphazardly. But the key question remains unanswered: who will watch the watchers? If bias audits are carried out by the police themselves, it is like putting the fox in charge of the henhouse. Independent external oversight, algorithmic transparency and mechanisms for appealing decisions all remain, for now, in the realm of good intentions.

For a Russian audience, this story serves as a reference point. Russian law enforcement is also actively adopting facial recognition and predictive analytics, yet there is practically no public discussion in Russia about bias in these systems. The British experience shows that even in a country with developed institutions of civil oversight and independent media, the problem of AI bias in policing remains unsolved. Acknowledging the problem is a necessary first step, but without concrete oversight mechanisms it risks remaining a nice gesture, followed by the quiet rollout of imperfect technology across the whole country.