Release time: 2026-08-02 12:24
AI alone won’t keep high profile figures safe
AI cannot replace human judgement in close protection and surveillance (CPS) services, according to new research published today.
The research, from the British Psychological Society's journal Legal and Criminological Psychology, concludes that given the challenges identified in the study, 'AI is not suitable to fully replace human analyses and assessments in CPS practice.'
The Dutch research is based on in-depth interviews with professionals from all organisations involved in CPS at a local or national level in the Netherlands, including police and counter-terrorism services. The findings were analysed by research academics with expertise in applications of AI and further explored at a meeting of experts in the field.
The findings highlight both the opportunities and the challenges of implementing AI in CPS services. Key concerns identified include:
Lack of appropriate data to train AI models
The ambiguity of information that professionals within CPS work with means that it can't always be labelled for model training
A tendency for large language models to 'hallucinate', that is to generate factually incorrect but seemingly plausible information
Unknowns about human-AI interaction, including concerns about bias and either under- or overreliance on AI-generated outcomes or advice, and the unique high-stakes, time-sensitive nature of close protection and surveillance
Whether people need protection against stalking or ex-partner violence, or public figures face threats from organised crime, non-state actors, or extremist groups, "CPS operates within a complex and constantly evolving criminal context", study author Isabelle van der Vegt says.
"For AI models designed to support decision-making in this context, it's therefore vital that models are trained with specialised knowledge. Also, continuous updating is critical to ensure the model remains aligned with emerging threats."
Responding to the findings, forensic psychologist and associate editor of the Journal of Applied Operational Intelligence Professor Brandon May said:
"It comes as no surprise to me that current AI systems are not suitable replacements for human decision-makers; in fact, our own research at the Center for AI in Policing has consistency shown that AI is not suitable for high-stakes tasks that require judgement, intuition and tacit knowledge.
"It is important to remind ourselves that large language models lack inference, understanding, and the forms of reasoning and cognition that underpin expert judgement. Although these models incorporate context through the conditional probability distribution governing next-token prediction, the concept of context remains different from the human capacity for intuition, knowledge, experiential understanding, and cognitive flexibility.
"CPS expertise is just not aligned to AI-facilitated technologies in their current state, not because the technology is not capable, but because the vulnerability to these tools remains with us."
