Health and Safety round-up: HSE statistics, the Building Safety Regulator, and HSE's line on AI

This week in UK Health and Safety

For a SHEQ director preparing a board paper, the recent regulatory calendar gave you three things worth citing: HSE's latest work-related fatality and injury statistics for Great Britain, a clearer picture of how enforcement resource is being deployed, and a published regulatory approach to AI. None of it is comfortable reading, and none of it points to a single fix. This round-up pulls the developments together and, where it helps, connects them to what real-time intervention actually does on a working site.

The headline numbers, and what sits underneath them

HSE's key figures for Great Britain, covering 2024 to 2025, set out the scale of the problem: work-related fatalities, non-fatal injuries reported through the Labour Force Survey and RIDDOR, and the sectors where the risk concentrates. Being struck by moving vehicles or objects and falls from height remain among the most common causes of serious harm. The annual statistics release for 2024 to 2025 carries the detail behind those headlines, and the DAC Beachcroft analysis reads them as the trends shaping HSE's strategic priorities for 2026.

The enforcement picture is more nuanced than the raw injury count. Construction News has reported that HSE enforcement notices have fallen to a five-year low. Set against a Register of convictions and notices that continues to record prosecutions for serious control failures, the pattern points to regulatory resource concentrating on a smaller number of high-impact cases rather than a broad softening. If you own a site P&L, that matters: the cost of getting caught on a serious control failure has not gone down.

What the 2026 regulatory calendar flags

The outlooks from Osborne Clarke and NFP both point to building safety, fire safety and asbestos as the pressure points to watch through 2026. For anyone tracking the regulatory calendar, those are the areas where duty-holders should expect scrutiny to sharpen. The practical read is straightforward: review your controls in those areas now, on your own timetable, rather than after an inspection.

HSE has also set out its own priorities in the KPI report for 2025 to 2026, which describes a programme of proactive inspection weighted towards health risks. If your site sits in a targeted sector, the takeaway is that inspection is being planned rather than purely reactive, and audit readiness is worth revisiting now rather than after a knock on the door.

HSE's line on AI in safety systems

HSE has published its regulatory approach to Artificial Intelligence. The message is proportionate rather than prohibitive: assess AI in machinery and workplaces on a risk basis, account for cyber threats, and favour transparent, explainable systems that work with existing controls rather than obscuring them. That framing matters for anyone specifying a vision-based system. It sets the expectation that AI in a safety context should support the existing hierarchy of controls, keep a human in the loop, and stand up to the same risk-based scrutiny as any other part of the safety case.

Where active safety fits the numbers

Most cameras on an industrial site are passive. They record footage that gets reviewed after an incident, when the value is evidence and investigation, not prevention. The recurring causes in the data, struck by moving vehicles, falls, PPE gaps, are precisely the moments where after-the-fact review comes too late. This is the gap active safety is built to close: detection that triggers a physical response in the moment, before the incident lands.

SAiFI, Dynamic Safety's active-safety platform, runs vision AI at the edge to detect hazards such as vehicle-pedestrian conflicts and exclusion-zone breaches and act on them in real time. It drives physical interventions, closing a dock gate, dropping a barrier, projecting a warning, raising an alarm, rather than logging an event for later. The design is offline-first, so it keeps working when connectivity drops, and it is configured per site with detection models, spatial zones, a rules engine and schedules that match how the site actually runs, rather than a generic model bolted on.

• •Real-time detection-to-action separates an active intervention from an after-the-fact alert.
• •Offline-first edge compute keeps interventions running through connectivity loss.
• •Per-site zones, rules and schedules reflect real operations, not a generic template.
• •Audit trails and cyber controls support the risk-based, explainable approach HSE expects from AI.
• •The platform works alongside banksmen, floor markings, mirrors and training; it does not replace them.

None of this displaces the fundamentals. The statistics are a reminder that people on the hundredth hour of a shift miss things, and that yellow lines and toolbox talks carry a site a long way but not the whole way. Active safety is the last line that catches what gets through. For a H&S manager, that is a point worth taking to the next safety committee; for a site director, it is an ROI argument grounded in the sectors HSE is already watching.