Professor Preeti Patel, Co-Director of the AI and Data Science Research Group, explores whether AI could improve public safety without needing to know who we are
Date: 21 September 2026
Facial recognition is becoming an increasingly visible part of public safety in the UK. At this year's Notting Hill Carnival, the Metropolitan Police expanded its use of Live Facial Recognition (LFR), with the technology contributing to more than 100 arrests. At the same time, its growing use has intensified questions around privacy, proportionality and public trust.
There are situations where identifying someone is clearly important. If police are looking for a particular wanted individual, for example, identity is central to the task. But this raises another question: what about the many public-safety situations where we don’t actually need to know who somebody is?
At airports, stadiums, festivals and transport hubs, risks can emerge through changing crowd dynamics. Congestion can build rapidly, movement patterns can shift unexpectedly and small changes in collective behaviour can develop into serious safety concerns. In these situations, the important information may not be who people are, but what is happening.
AI can analyse movement trajectories, crowd density, posture, direction of travel and interactions between people. Instead of asking “who is this person?”, a system could ask, “what is happening here?”.
It might detect an unusual build-up of people, movement against the main flow, sudden dispersal or rapidly increasing crowd density. This could provide public-safety teams with an additional layer of situational awareness, helping them recognise developing risks earlier.
Identity-agnostic surveillance
This idea is influencing research within the AI and Data Science Research Group.
Our work is exploring an approach described as identity-agnostic surveillance: designing AI systems around the principle that identity should not be acquired or inferred unless it is genuinely necessary for the safety objective.
Rather than assuming that better surveillance requires better identification, we are asking how much useful safety information can be gathered from non-identifying characteristics such as movement, posture, spatial relationships and group interactions.
The distinction matters: identifying a wanted individual requires identity, while detecting dangerous crowd density, an emerging bottleneck or unusual changes in movement may not.
This suggests that a third possibility in the familiar debate between public safety and privacy could be systems that provide useful situational intelligence while deliberately minimising their dependence on identity.
Building systems people can trust
Removing identity does not automatically make surveillance ethical or reliable. Human behaviour is highly dependent on context. A crowd celebrating a goal behaves very differently from passengers moving through an airport, and AI must distinguish normal behaviour from situations that genuinely require attention.
There are also questions of accountability. If AI indicates that a situation is becoming high risk, operators need to understand why, decide whether intervention is appropriate and remain responsible for the decision.
As AI plays a greater role in public safety, the focus should not simply be on what the technology can do, but on how it can be designed to serve a clear and proportionate purpose. The real opportunity lies in developing systems that enhance situational awareness while placing privacy, transparency and human judgement at the heart of their design. Building public trust will depend not on making surveillance ever more powerful, but on demonstrating that AI can be both effective and responsible.
Professor Preeti Patel is the Co-Director of the AI and Data Science Research Group.