Skip to main content
Blog·Retail trends

Protecting shop workers: the new law, and what AI on CCTV can and can't do

The Crime and Policing Act 2026 made assaulting a retail worker a standalone offence. The law raises the stakes on evidence. Here's what AI on existing cameras can do for staff safety, and where it stops.

SESam Erpik · Co-founder & CTO7 min read

For years, violence against shop workers was treated as part of the job. That changed this spring. The Crime and Policing Act 2026, which received Royal Assent on 29 April 2026, created a standalone criminal offence of assaulting a retail worker in England and Wales, and removed the £200 threshold that had let so much shop theft go uninvestigated.

The law did not appear from nowhere. The BRC's 2025 Retail Crime Survey recorded violence and abuse against retail workers rising more than 50% in a single year, and 340% since 2020, to more than 2,000 incidents every single day, the highest the survey has ever recorded.

What the new offence changes for retailers

A standalone offence does two things. It signals that an assault on a shop worker is its own crime, not a footnote to a theft. And it raises the practical importance of evidence, because an offence the police and CPS can actually charge is an offence that needs a clean, forward-ready trail behind it.

That second point is where most retailers are weakest. A standalone offence is only as useful as the evidence that supports a charge. A blurry phone photo of a monitor does not get a charging decision. A timestamped, multi-camera, audit-logged pack does.

The reframing is the opportunity: the same incident workflow that forwards a theft pack to police can forward an assault pack. The Act makes the second one matter as much as the first.

Where AI on existing cameras genuinely helps with staff safety

Camera-based AI is not a guard. But on the cameras a store already has, it can do three things that bear directly on staff safety:

  • Faster awareness. Restricted-zone and after-hours rules can page a manager the moment something fires, so a lone or near-lone worker is not the only person who knows an incident is unfolding.
  • Lone-worker context. On a thin night shift, the system becomes the second pair of eyes, the difference between 'no one is watching' and 'the system is watching', which is as much about how staff feel as what actually happens.
  • Prosecution-grade evidence. When an incident does happen, multi-camera tracking and an append-only audit log assemble the forward-ready pack the new offence now depends on, automatically, instead of a manager scrubbing footage afterwards.
app.quantumeye.io/events/safety

Safety Events

Restricted zones, PPE and occupancy

Illustrative
Open alerts
2
Needs review
Resolved today
7
Across 4 zones
Avg resolution
4m
−1m vs last week
Zone setup
Northgate · floor plan
ENTRANCEStockroom · CAM-09
Restricted zoneCameraFixtures
Safety alerts
Today
  • Restricted-zone entry
    Stockroom · CAM-09
    14:12Open
  • PPE missing — hi-vis
    Loading bay · CAM-11
    13:48Open
  • Max occupancy reached
    Fitting rooms · CAM-06
    12:30Resolved
  • Loitering
    Tills · queue line · CAM-02
    11:54Resolved
  • Restricted-zone entry
    Stockroom · CAM-09
    09:21Resolved
A restricted-zone or safety rule firing for human review, with the clip and context already attached. Illustrative.

Where it stops, and that matters

Being honest about the limits is the whole point of a product in this space. AI on CCTV does not de-escalate a confrontation, does not put a person between a worker and an aggressor, and does not replace a panic alarm, a safe-working policy, or trained staff. It is a detection and evidence layer, not a physical-security one.

And every consequential action stays human-reviewed. A camera flag is the start of a process a person owns, not an automated accusation. That is non-negotiable in a surveillance product, and doubly so when the subject of a clip might be a customer having the worst day of their life rather than an offender.

What we would tell a retailer to do now

  1. Treat the new offence as an evidence problem as much as a policy one. Make sure an assault on a colleague produces the same forward-ready pack a theft does.
  2. Map your lone-working windows, the night shift, the early open, the single-staffed hour, and decide what 'someone else knows immediately' looks like for each.
  3. Layer AI on the cameras you already own for awareness and evidence, on top of, never instead of, alarms, training, and a safe-working policy.
How QuantumEye handles safety and restricted-zone events
Rule-based alerts, after-hours cover, human-reviewed, evidence-grade output

Get the monthly brief.

One email a month. The post-of-the-month, the retail-trends summary, and one customer-success snippet. No sales pitches, no event invites. Opt out in one click.