Malaysia's CDM 2024 regulations require principal contractors to "plan, manage, and monitor the construction phase." Clients must "take reasonable steps" to ensure their appointees comply. These are continuous obligations -- not a form you fill once and file.
The penalty structure is real: up to RM500,000 in fines, 1 year imprisonment, and stop-work orders that can freeze an entire project (P.U.(A) 147/2024). Falls from height still account for 32% of all workplace fatalities in Malaysia (Zermane et al., SEEU Review 2024).
The regulations are clear about what is required. The harder question is how sites actually deliver continuous monitoring in practice.
How most sites monitor safety today
The standard approach on a Malaysian construction site:
- Periodic walkabouts by the Safety and Health Officer (SHO)
- Manual PPE spot-checks
- Paper-based inspection checklists
- Toolbox talks before shift starts
This was acceptable when safety management was voluntary under the 2017 Construction Industry Guidelines. Under CDM 2024, the standard has shifted from "did you have a safety plan?" to "can you demonstrate that monitoring was continuous?"
Where the coverage breaks down
Malaysian regulations do not prescribe a specific ratio of safety officers to workers. 1 SHO may be responsible for an entire construction site. With approximately 1.4 million workers across 132,272 registered contractors (OpenDOSM), the coverage gap is structural, not a failure of individual diligence.
A single safety officer cannot simultaneously monitor all zones of a large construction site. Workers can remove PPE after the inspector passes. Paper records can be backdated. Toolbox talk attendance does not prove on-site compliance.
On sites with hundreds of workers -- many of them foreign workers with limited English or Bahasa Malaysia -- manual monitoring cannot cover every zone at every hour.
Research from Malaysian universities confirms this: the gap between documented safety management systems and actual worker behaviour on site is a persistent, well-documented problem (Zermane et al., 2024; IOP Science, 2022).
Why the gap matters legally under CDM 2024
CDM 2024 requires demonstrable compliance. If DOSH investigates after an incident, "we had a safety plan" is weaker than "here is a timestamped, continuous monitoring record showing what was happening on site."
DOSH can issue stop-work orders -- and they do. In 2019, DOSH Pahang issued stop-work orders to 16 construction sites in a single enforcement operation (DOSH Archive). In April 2025, Selangor DOSH issued a stop-work order after a fatal crane incident at a Shah Alam construction site (Yahoo News Malaysia).
A stop-work order on a RM150 million project costs far more in delays than any fine.
How AI safety monitoring closes the gap
Computer vision systems, running on existing CCTV infrastructure, address the specific monitoring challenges that CDM 2024 creates.
PPE compliance monitoring
Computer vision detects whether workers are wearing required PPE -- hard hats, high-visibility vests, safety harnesses, gloves, safety boots -- in real time across every camera on site. Academic research confirms that YOLO-based PPE detection systems achieve over 90% accuracy in controlled settings (PeerJ, 2022), and Malaysian universities are actively developing and testing these systems for local construction environments (UTM Jurnal Teknologi).
Instead of periodic manual spot-checks, the system monitors continuously. When a worker enters a zone without the required PPE, the system flags the violation immediately -- not at the next scheduled inspection.
Zone monitoring and restricted area detection
CDM 2024 requires principal contractors to control site access. AI video analytics can define virtual restricted zones around excavations, crane operating radii, chemical storage areas, and elevated work platforms. When a worker enters a restricted zone without authorisation, the system generates an alert.
Falls from height are the leading cause of construction fatalities in Malaysia. A zone monitoring system that detects workers near unprotected edges or in elevated areas without visible harnesses runs 24 hours a day. A safety officer does not.
Unsafe behaviour detection
Beyond PPE and zone violations, AI systems can detect dangerous behaviours: workers operating under suspended loads, using mobile phones in hazardous zones, or working at height without proper fall protection (Automation in Construction, 2025).
The compliance evidence trail
Every detection generates a timestamped record. This is the critical difference for CDM 2024 compliance: instead of paper checklists that prove only that someone walked the site at a point in time, AI monitoring produces a continuous, verifiable evidence trail of safety conditions across the entire site.
When DOSH investigates, this data demonstrates that the principal contractor was actively monitoring -- not just documenting after the fact.
CIDB Construction 4.0 and where digital safety fits
Malaysia's Construction Industry Development Board (CIDB) has published a Construction 4.0 framework that explicitly promotes AI, automation, smart monitoring, and data analytics for the construction industry (CIDB SMART Portal). In October 2025, CIDB launched the Construction Digital Maturity Index (CDMI) -- the first tool to benchmark contractors' digital readiness (The Edge Malaysia, October 2025).
Digital monitoring is moving from optional to part of how compliance is demonstrated.
With the construction sector growing 15% in real terms in 2024, major projects like MRT3 and the Penang LRT (RM8.3 billion civil works) ramping up, and CDM 2024 enforcement maturing, the gap between paper-based safety management and demonstrable, technology-supported compliance will widen.
How Hypernology can help
HyperQ AI Safety deploys VLM-powered (Vision-Language Model) safety monitoring using your existing CCTV infrastructure. The system is operational in approximately 1 hour -- no new cameras required.
What it monitors:
- PPE detection: hard hats, high-visibility vests, safety harnesses, gloves, safety boots
- Zone monitoring: virtual restricted zones around excavations, crane areas, elevated platforms, and chemical storage
- Unsafe behaviour detection: workers in prohibited zones, working at height without fall protection, posture-based hazard detection
- Fire and smoke detection: context-aware AI that distinguishes between a welding flame and an actual fire, reducing false alarms in construction environments where controlled heat is normal
HyperQ AI Safety also integrates with our Smartband for worker biometric monitoring -- body temperature, heart rate, oxygen saturation (SpO2), and blood pressure -- with alerts sent directly to the worker via vibration and to the control system simultaneously.
Every detection generates a timestamped record that supports your CDM 2024 compliance documentation.
HyperQ AI Safety is built for construction sites in Malaysia and the broader APAC region. We focus specifically on AI machine vision for safety monitoring -- that is what we do, and that is all we claim to do. For companies operating in Singapore, the Enterprise Development Grant (EDG) may cover a significant share of qualifying AI implementation costs.
If your project involves construction site safety monitoring and you want to explore how AI can support CDM 2024 compliance, talk to us about your site.
Sources
- DOSH CDM 2024 Gazette, P.U.(A) 147/2024: intranet.dosh.gov.my
- Zermane et al. (2024), "Statistical Analysis of Occupational Fatal Accidents for Risk Assessment," SEEU Review 19(1): reference-global.com
- IOP Science (2022), Safety Management Systems vs Actual Behaviour: iopscience.iop.org
- DOSM Construction Statistics Q2 2024: dosm.gov.my
- DOSH Stop-Work Orders, Pahang (2019): intranet.dosh.gov.my
- PeerJ, YOLO-based PPE Detector (2022): peerj.com
- UTM Jurnal Teknologi, PPE Detection in Malaysian Construction: journals.utm.my
- Automation in Construction (2025), Unsafe Behaviour Detection: accedacris.ulpgc.es
- CIDB Construction 4.0 Guide: smart.cidb.gov.my
- The Edge Malaysia, CIDB CDMI Launch (October 2025): theedgemalaysia.com