Work & Safety / AI news for Malaysia
From the archive · Event date 27 May 2025
UPM discussed AI for workplace safety. No system was announced
The speakers said digital tools could detect risks earlier and monitor workplaces continuously, while humans and basic PPE discipline remained essential. UPM's report described a knowledge-sharing event, not an operational AI rollout.
In brief
- UPM held a knowledge-sharing session with Malaysia's Department of Occupational Safety and Health during its occupational safety and health week on 27 May 2025.[1]
- A DOSH Selangor speaker said AI and digitalisation could support early risk detection, accident reduction and systematic workplace monitoring, while emphasising that humans must remain in control.[1]
- UPM's report did not identify a deployed AI system, workplace trial, dataset, detection result, accident reduction or independently measured outcome.[1]
UPM hosted a real safety discussion, but did not launch an operational AI monitor
Universiti Putra Malaysia brought occupational-safety and technology speakers together on 27 May 2025 to discuss how digitalisation and artificial intelligence could strengthen workplace safety and health. The session was held with the Department of Occupational Safety and Health, known as JKKP or DOSH, as part of UPM's workplace-safety week.[1]
Dr Mokhtar Zamimi Ranjan, head of investigation and prosecution at DOSH Selangor, said AI and digital tools could support earlier risk detection, fewer accidents and more systematic monitoring. He also warned against treating technology as the decision-maker: digital systems may operate continuously, but people remain responsible for balanced and ethical use.[1]
The event established a direction for discussion, not a technology deployment. UPM's first-party report did not name an AI product, sensor network, pilot site, data source, alert threshold, evaluation period or measured change in workplace incidents. The safest reading is that UPM examined a potential operating model rather than announcing a proven safety system.[1]
Early risk detection starts with the signals a workplace can actually observe
A useful workplace-safety system needs a defined input. Depending on the environment, that could be equipment readings, noise exposure, temperature, vehicle movement, video, near-miss reports, maintenance logs or worker observations. AI can help identify unusual patterns, but it cannot detect a hazard that was never measured or reported.[1]
Mokhtar's emphasis on continuous digital operation points to one benefit: machines can monitor selected signals outside ordinary inspection rounds. Yet continuous collection is not the same as continuous protection. Every alert needs a responsible person, response time, escalation route and record of what action followed.[1]
The report did not say which hazards UPM had prioritised or whether any model had been tested. Without that detail, readers should not infer computer vision, predictive maintenance or wearable monitoring from the general AI label. Each approach has different data, accuracy and worker-impact requirements.[1]
The speakers treated AI as a complement to people and PPE—not a replacement
Datuk Ts Is Mohd Faiz Zulkifli of My Lab Scientific told the event that technology must be accompanied by consistent safety practice and disciplined use of personal protective equipment. He described AI as a strong tool for safety management, but one that should empower people rather than replace them.[1]
That hierarchy is important. Employers should first remove or reduce hazards through workplace design, engineering controls and safe processes. PPE is a final defensive layer, while AI may support several layers by detecting patterns, reminding workers or helping supervisors focus attention. An AI alert cannot make unsuitable equipment safe or compensate for missing training.[1]
The UPM account also quoted a warning about rising workplace accidents in 2023, with Selangor said to record the highest number particularly in manufacturing and services, and hearing loss linked to continuous noise exposure named among major occupational diseases. The article did not link the underlying statistical table, so those figures should be checked against the official dataset before being used as a deployment baseline.[1]
A credible pilot would measure safety improvement and false confidence together
The basic performance question is whether the tool finds a meaningful hazard earlier than existing practice. A pilot should record true alerts, missed hazards, false alarms, response time and the severity of each event. A system that produces too many weak warnings can be ignored, while one that misses rare high-impact risks can create dangerous confidence.[1]
Workforce evidence matters too. Workers and safety representatives should know what is being monitored, how an alert affects them and how to challenge a wrong conclusion. Feedback can reveal blind spots that a technical accuracy score misses, including whether people change reporting behaviour because they feel watched or unfairly judged.[1]
UPM or a participating employer could then publish a before-and-after scorecard: near misses, verified hazards, corrective-action time, incident frequency, exposure levels, false alerts and training completed. That would convert a useful awareness event into evidence that Malaysian employers can compare and learn from.[1]
Why Malaysia should care
Malaysian employers face a practical question: where can AI improve hazard detection without creating false confidence? UPM's event correctly kept people, discipline and protective equipment in the picture. The next step would be transparent workplace trials with measurable safety outcomes and worker participation.
Workers
Digital tools may detect selected hazards between ordinary inspections, but should not weaken existing controls or worker voice.[1]
Practical move: Ask what is monitored, who reviews alerts, how errors are challenged and whether PPE and training remain funded.
Safety teams
AI may help prioritise attention when the input, hazard and escalation pathway are clearly defined.[1]
Practical move: Run bounded pilots with baseline data, human verification, false-alarm limits and documented corrective actions.
Employers
A technology label does not prove accident reduction or fulfil ordinary safety responsibilities.[1]
Practical move: Publish measurable safety outcomes and keep engineering controls, safe processes, training and PPE ahead of automation claims.
What Malaysians can do now
- Name the exact workplace hazard, data source, AI output and responsible human decision before starting a pilot.
- Measure missed hazards, false alarms, response time, corrective action and incident or exposure change against a baseline.
- Explain worker participation, monitoring boundaries, appeal routes and how existing controls and PPE remain protected.
What we still do not know
The potential was discussed; no workplace AI system or measured safety result was reported.
- Whether UPM or a participating organisation later deployed a named AI workplace-safety pilot.
- Which hazards, sensors, datasets, workplaces and workers would be included in any operational system.
- Whether a trial improved verified hazard detection or reduced incidents without unacceptable false alarms or worker harm.
Sources
- 1.UPM Terap Pendigitalan dan AI dalam Keselamatan Pekerjaan Universiti Putra Malaysia, 27 May 2025
