Artificial intelligence (AI) is rapidly transforming the functioning of police forces and the criminal justice system. From the analysis of digital evidence to the automatic drafting of documents or the identification of criminal patterns, these technologies promise to improve efficiency and reduce the workload of professionals. However, a new study led by Northumbria University warns that the implementation of these tools is advancing more rapidly than the necessary control, oversight, and regulatory mechanisms to ensure safe, transparent, and responsible use.

The study, developed over four years as part of the PROBabLE Futures project, constitutes the most comprehensive map created to date on the use of AI in the criminal justice system of England and Wales. Researchers have identified a total of 70 artificial intelligence tools that are already operational, in testing phase, or in development. Of these, 27 are already being used operationally, while 34 are in pilot phases. More than half have been developed by private companies, a fact that also highlights the growing role of the technology sector in this area.
The detected applications are very diverse. They include automatic transcription of statements, assisted drafting of reports, classification of calls to emergency centres, analysis of large volumes of information to detect criminal patterns, identification of vulnerable individuals, management of digital evidence, drafting of legal documents, and facial recognition systems. There are also tools aimed at the well-being of officers, capable of identifying indicators of stress or psychosocial risk.
Researchers acknowledge that many of these applications provide real benefits when implemented correctly. Repetitive and administrative tasks can be automated, allowing professionals to dedicate more time to activities that require human judgement. Similarly, automated data analysis can facilitate faster investigations and help detect relevant information that might be overlooked in a manual review.
However, the main message of the report is that these advantages are only sustainable when the tools have been designed to address specific problems, have undergone rigorous testing, and operate under robust oversight mechanisms. According to the authors, this situation is not yet widespread, and in many cases, the speed of adoption exceeds the capacity of institutions to assess risks and establish clear governance standards.
One of the highlights of the study is the criticism of the concept of human in the loop, that is, the idea that a person always reviews the decisions or recommendations generated by AI before they have practical effects. In theory,
this mechanism should ensure that the final responsibility continues to rest with the individuals.
However, researchers observe that this oversight is often more formal than effective. When a tool demonstrates a high level of accuracy, users tend to trust it excessively and stop reviewing its results with the same attention. This creates a false sense of security: errors are infrequent, but when they occur, they are more likely to go unnoticed and can lead to significant consequences, especially in sensitive areas such as police investigations or judicial processes.
The report also warns of another emerging risk: the connection of various AI systems within the same process. If the information produced by one tool automatically feeds into another, any initial error can be transmitted and amplified throughout the entire decision chain. This phenomenon, known as «AI system chaining», has still been little studied, but the authors believe it deserves priority attention.
To address these challenges, the study formulates 26 recommendations aimed at public administrations, law enforcement agencies, the judicial system, technology providers, and the scientific community. Among the proposed measures are the conduct of independent assessments of all AI tools, the creation of public registries to know which systems are in use, the establishment of more stringent standards for their acquisition and validation, better training for professionals, and specific research on the risks arising from the interaction between different artificial intelligence systems.
In short, the study does not question the usefulness of AI in the police field nor does it advocate for halting innovation. On the contrary, it concludes that these technologies can bring great value when used responsibly and transparently. The real challenge is to ensure that technological development is accompanied by a robust framework of governance, oversight, and accountability. Only in this way will it be possible to harness the benefits of artificial intelligence without compromising fundamental rights, the quality of investigations, and the trust of citizens in security institutions.
_____
Aquest apunt en català / Esta entrada en español / Post en français








