It is 23:47 on Tuesday. The night shift has been running for three hours. On the security panel, everything is green. The cameras record every corner of the plant, just as they have been doing for the past four years.
At 00:12, an operator enters the loading zone without a helmet. Nobody sees it. The camera records everything, perfectly in focus, with a flawless timestamp and resolution.
Three days later, reviewing footage after a minor incident, the safety manager finds it. The video was there. The camera did its job.
The safety system did not.
This scene repeats itself every week at plants across Spain. The problem is not the cameras — it is what we do (or do not do) with what they record. If you work in industrial safety, this post is for you.
The camera that sees everything… and does nothing
Installing cameras at an industrial plant has become the first reflex before any safety audit. It is understandable: they are visible, they document, they give a sense of control.
But there is a fundamental difference between recording and preventing. A passive camera does the former. Prevention requires something else.
According to data from the Ministry of Labour Work Accident Statistics, in 2024 more than 628,000 workplace accidents resulting in sick leave were recorded in Spain — 10.4% more than the previous year. An upward trend that has gone unreversed for more than a decade.
Do those plants have cameras? Most of them, yes. Are they being used effectively? That is the question few people ask.
The real cost of reviewing footage (and why no one does it properly)
The traditional camera surveillance model has one obvious bottleneck: the human eye.
An OHS officer reviewing footage cannot maintain constant attention for more than 20 consecutive minutes without their detection ability dropping significantly. That is biology, not lack of professionalism.
Add to that the fact that a medium-sized plant may have 20, 40 or more active cameras. Reviewing all footage from a single shift exhaustively would take longer than the shift itself. In practice, recordings are only reviewed after something has already happened.
The result: the camera becomes a post-accident investigation tool, not a prevention tool. The difference between the two is not semantic — it is the difference between preventing an accident and documenting it.
The blind spots no audit ever sees
There are risks that periodic reviews simply do not capture. Not because the officers are negligent, but because they are invisible by their very nature.
The most common ones according to specialist studies:
• Occasional PPE non-compliance: the operator who removes their helmet "just for a moment". It is not bad faith — it is protection fatigue. It happens hundreds of times a day at an active plant.
• Access to restricted zones outside working hours: especially during night shifts, when human supervision is thinnest.
• Ergonomic risk behaviours: incorrect loads, forced postures. EU-OSHA estimates that musculoskeletal disorders account for more than 60% of occupational diseases.
• Early signs of fatigue: changes in movement pace, unplanned pauses, disruptions to repetitive tasks.
None of this shows up in a monthly audit. All of it happens, at varying frequencies, every single day.
What changes when the camera starts thinking
Computer vision is not magic — it is a shift in what we can ask a camera to do.
Instead of recording to review later, an AI system analyses video in real time and generates automatic alerts when it detects a risk situation: a worker without a helmet, a person in a restricted area, a movement pattern associated with fatigue.
It does so continuously, without tiring, without bias, without "giving the benefit of the doubt" to a trusted colleague. And it documents everything — not to investigate accidents, but to demonstrate compliance and detect patterns before they escalate.
This is exactly what Safe does: turning the cameras you probably already have installed into an active prevention system. No construction work, no infrastructure changes, no one staring at a screen.
The OHS officer does not disappear — quite the opposite. They are freed from manual review to focus on training, protocols, and decisions. Safe gives them the objective data; they decide what to do with it.
Where do you start?
The most common question when someone sees Safe in action is not "does it work?" — it is "how long does it take to implement?"
The honest answer: it depends on the starting point, but far less than is usually assumed. At plants with IP cameras already installed, the first use cases can be live within days.
What does require time — and is worth giving — is defining what you want to detect, in which zones, with what alert thresholds. That part is not done by the AI: it is done by the security team that knows the plant.
The technology provides the eyes. You provide the judgement.
Your cameras are already there
If you have cameras installed at your plant, you already have the infrastructure for an active safety system. All that is missing is giving them intelligence.
The difference between a plant that documents accidents and one that prevents them is not in the number of cameras — it is in what happens between the image and the alert.
Want to see how Safe would work at your plant? We can talk without obligation in under 30 minutes.