AI is changing the world at an unprecedented pace. Used in everyday life, the technology has now been used by a British police officer not to correct grammar or spelling mistakes but to produce evidence where none existed.
The police officer in Derbyshire has been suspended on suspicion of perverting the course of justice, and the force is liaising with the Crown Prosecution Service to see if ongoing criminal cases have been affected.
While police forces have increasingly begun using AI to speed up processing and improve efficiency, concerns have now emerged about its potential misuse. Earlier last week, PolicingAI was launched by Sarah Jones, the policing minister, with the aim of harnessing the technology to strengthen the fight against crime.
Backed by a £75m ($101m) budget, the program will run large-scale pilots to help officers triage, disclose and summarize digital evidence, which is among the most time-consuming parts of any investigation.
AI is also extensively used by Palantir, the American defense technology company that has an office in the UK and provides AI investigation tools to 11 British police forces. Its technology helped detectives in Bedfordshire process digital evidence in a case involving a criminal gang that had stolen £800,000 ($1.1m).
AI Error at Heart of Football Fan Ban
The risks associated with the growing use of AI in policing were highlighted earlier this year by a separate controversy involving West Midlands Police.
An official review found that inaccurate information generated by Microsoft's Copilot AI was used in intelligence assessments that helped justify a ban on Maccabi Tel Aviv supporters attending a Europa League match against Aston Villa in Birmingham. The report included references to incidents that had never occurred, with the force later admitting that some of the information originated from an "AI-generated hallucination".
The case triggered a political storm after senior officers initially denied that AI had been involved. Chief Constable Craig Guildford later apologized for providing misleading information to MPs.
A subsequent review concluded that the threat posed by Maccabi supporters had been exaggerated and that flawed intelligence had played a role in the decision-making process. The findings prompted Home Secretary Shabana Mahmood to say she had lost confidence in Guildford, who retired shortly afterward.
Concerns were also raised that the police had deliberately used AI to seek intelligence which could be used to ban the Maccabi fans, after local Muslim politicians lobbied them to do so. The police intelligence report included numerous false claims about the Maccabi fans, while playing down the threat posed to them by local Muslim activists.
Wider Concerns About AI
Policing is not the only field where AI has been misused. A recent investigation by Wired magazine found that Iran had been using Google Gemini to alter satellite images. Long seen as a gold standard of visual intelligence, satellite imagery was exploited by Iran in an attempt to develop war propaganda. Images published in the Tehran Times appeared to show before-and-after images of a US base in Bahrain, with a US radar supposedly destroyed in the latter.
Similar techniques were used to erase evidence of US and Israeli strikes on Iranian air force bases and oil refining facilities, removing bomb craters and burn marks from images taken from commercial satellites. No longer can it be assumed that a published satellite image is automatically reliable.
Although all these cases raise questions about the potential misuse of AI, in every case they were rapidly discovered. That indicates that fears of widespread misinformation and deception through the use of AI have so far failed to materialize.
As with other new technologies, adoption will require new training and oversight mechanisms, as well as cultural understanding around its potential and limitations.
In the case of West Midlands Police, it was revealed that officers had access to AI tools but had received no instruction in their use because of the cost. Instead, only office-based support staff were trained to operate them. Experts criticized the arrangement, arguing that the cost of such training should be minimal and that clear guidance was needed.
As AI becomes increasingly prevalent in society, such events will cease to surprise and will instead be part of ordinary life, with guidelines and verification evolving to tackle new challenges.
