Cameras and Analytics at Work: Keeping Safety from Becoming Surveillance

Where safety ends and surveillance begins

The line erodes gradually. A system is installed to prevent theft or reconstruct injuries; months later the same cameras are checking break times, picking speed, or error rates. An analytics feed meant to flag a forklift driver without a helmet gets reprogrammed to count minutes of idle time. This "function creep" is what most often turns a safety measure into surveillance.

The evidence cuts both ways. The U.S. Government Accountability Office (GAO) found that cameras, microphones, location trackers, and wearables can increase a sense of security and even flag emerging health problems — but they can also raise stress and anxiety and even raise injury risk when they push people to move faster to hit productivity targets. The question is rarely "can we install a camera" but "for what single purpose and under what constraints."

It also helps to separate levels of intrusion. A camera covering a gate or a racking aisle with hazardous machinery controls a zone; a camera trained on one employee all day observes a person. The closer a system moves toward an individual worker, the heavier the justification required and the easier it becomes to cross the line.

Where the law draws the line

In the EU, video footage is personal data under the GDPR, and the European Data Protection Board (EDPB) stresses that employees generally do not expect to be monitored at work, so the balance of interests must be weighed with particular care. A system must be adequate, relevant, and not excessive; each camera needs a specific written purpose, and any disclosure of footage to third parties is a separate processing activity needing its own legal basis.

In the UK, the UK GDPR and the Data Protection Act 2018 apply, and the Information Commissioner's Office (ICO) advises employers to monitor lawfully and fairly, choosing the least intrusive means and not intruding into workers' private lives. Homeworking raises the stakes because expectations of privacy at home are higher. Note that the ICO's monitoring guidance is under review following legislative change, so check the current version.

A separate layer is the EU AI Act (Regulation (EU) 2024/1689). Since February 2025, emotion recognition in workplaces and education is prohibited in the EU, as is biometric categorisation used to deduce protected characteristics. Tools for employment, worker management, and access to self-employment are classified as high-risk, bringing requirements on risk assessment, data quality, logging, documentation, transparency, and human oversight. As of mid-2026 these employment obligations are expected to apply from 2 December 2027, but the timeline has shifted — verify official Commission updates.

These are general orientation points, not legal advice, and rules differ by jurisdiction. Before designing a system, review national law and relevant collective agreements, and confirm the project with qualified counsel.

Designing a program that stays on the safety side

Start with the problem, not the technology. Describe the specific risk — a fall hazard, vehicle strike, unauthorized entry, warehouse theft — then test whether a non-video control (guarding, lighting, sensors, interlocks, procedure) can meet it. Cameras and analytics should appear only where gentler measures fail.

Give each camera a single written purpose and document necessity and proportionality; in the EU, large-scale monitoring usually requires a data protection impact assessment (DPIA). Draw a placement map: toilets, changing rooms, break areas, first-aid and medical rooms are off-limits, and do not keep a camera trained permanently on a fixed workstation.

Set the technology's own limits: short retention with automatic deletion, access controls and viewing logs, encryption, and restrictions on export. Define precisely which events analytics may flag — "a person entered the hazard zone," for instance — and what it must never do: emotion recognition, face-based identification, or biometric categorisation.

Write into policy that data collected for safety is not used for disciplinary decisions, performance ratings, or pay calculations unless there is a separate lawful basis and notice. Put that boundary in writing rather than relying on a supervisor's discretion.

• One written purpose per camera and per analytics algorithm
• Exhaust non-video controls first: guarding, lighting, sensors, procedures
• Forbidden zones: toilets, changing rooms, break areas, medical rooms
• No permanent camera over a fixed workstation
• Short retention, access logs, encryption
• Analytics flag only safety events — never emotions or biometric categories
• Separate "safety" from "performance and discipline" in written policy

Keeping analytics from becoming discipline and bias

Video analytics earns its place by turning streams of frames into events: someone dropped their hard hat, someone entered the crane swing zone, a fall occurred. Yet the same data can produce unfairness. Drawing on interviews and 122 studies, GAO found that when tools rely on flawed productivity benchmarks or ignore the full range of duties, workers are more likely to receive low evaluations, penalties, reduced pay, or termination.

Bias is a particular risk. Metrics that ignore individual circumstances — for example, that a worker with a disability may need more time — can disadvantage whole groups. GAO also notes that many workers and unions worry surveillance deters organizing and leaves people feeling constantly watched, which erodes morale and productivity.

For that reason, every analytics alert should pass through human review, and workers should have a channel to contest a decision. The algorithm can suggest "possible incident," but a person makes the finding and the decision. That reduces both errors and the sense that an impersonal machine is in charge.

Audit the system quarterly for drift: have new scenarios crept in, is data flowing to HR without a basis, has retention lengthened? The earlier you catch creep, the cheaper it is to reverse.

Trust is built with people, not just policy

Transparency is the strongest single tool. Inform workers before activation — where cameras are, what analytics sees, who has access, how long footage is kept, and how to contest a decision. Visible signage is mandatory. When people understand a system protects them from injury rather than watching their breaks, anxiety falls.

Where the framework provides for it, involve worker representatives and unions and agree the monitoring scheme before launch. That is not only legally sound in many jurisdictions but practical: the workforce itself will tell you where a camera helps and where it creates tension.

GAO's research shows the effect on mental health depends heavily on employer practice — in particular, how honestly the organization explains what it collects and why. Cameras for safety that are discussed openly and genuinely protect are received differently from covert monitoring with unclear purposes.

Remember the goal: fewer injuries and a real sense of protection. If incident counts fall but turnover and stress complaints rise, the system has stopped being a safety tool and has become a source of distrust.