How Can Brazilian Plants Monitor NR-12 and NR-35 Risks With AI?
Updated: 13 hours ago

“Quick AI-Powered Insights on the Topic— Freshly Updated!”
|
Brazil recorded 806,011 occupational accidents and 3,644 work-related deaths in 2025, the highest annual totals in its 2016–2025 series, according to the Ministry of Labour and Employment's official accident study. These figures raise an important question for industrial safety in Brazil:
How can plants identify unsafe machinery interactions and work-at-height hazards before they result in injuries?
Brazil's NR-12 and NR-35 standards establish requirements for machinery safety and work at height, respectively. Alongside these established controls, AI video analytics for workplace safety can help plants monitor selected visible hazards through suitable CCTV cameras, generate alerts and document events for safety teams to investigate.
The opportunity is not to replace inspections, guards, interlocks or permits. It is to improve visibility into what happens on the production floor between scheduled safety checks.
Why is industrial safety in Brazil under pressure?
Industrial safety in Brazil is under pressure because reported work accidents in Brazil have climbed steadily since 2020 and reached a record in 2025. Fenacon reports 806,011 accidents in 2025, up 65.8% from the 2020 low of 486,303, with deaths up 60.8% over the same period. Across 2016 to 2025, the MTE study counts 6.4 million accidents, 27,486 deaths and more than 106 million working days lost (Portal Contábeis). The incidence rate relative to formal employment fell over the series because employment grew, but the absolute count still rose.
For plant managers, the occupation ranking is the closer signal. Production line feeders (alimentadores de linha de produção) rank second among occupations for both accidents and deaths, behind nursing technicians and truck drivers respectively, according to Bra1, citing Secretaria de Inspeção do Trabalho data drawn from INSS notifications.
What conditions put production line workers at risk?
Three operating conditions recur in machine and height work. These are common patterns in industrial safety practice, not findings of the MTE study.
Hands near moving parts. Feeding material, clearing jams and cleaning put hands and arms inside danger zones, sometimes while the machine is still energised.
Guards removed or bypassed. A guard taken off to speed a changeover or keep up output is easy to leave off afterwards.
Short, non-routine tasks at height. Clearing a blockage on a mezzanine or inspecting a conveyor gantry is often done quickly, without a permit, anchor point or harness.
What does NR-12 require for machine safety?
NR-12 sets minimum requirements to prevent accidents in the design and use of machines and equipment across all economic activities (item 12.1.1). Protection follows an order: collective measures first, then administrative or organisational measures, then individual protection (items 12.1.7 and 12.1.8).
The items most relevant to monitoring are:
Safety systems on danger zones (12.5.1): Danger zones need fixed guards, movable guards with interlocks, or safety devices. Fixed guards can only be removed with tools; movable guards must be paired with an interlock (12.5.4).
Moveable guards and Interlocking (12.5.6 to 12.5.8): Where access to a danger zone is required more than once per work shift, the protection must be movable and associated with the appropriate interlocking device. Guard locking is required where opening the guard could allow access before the risk has been eliminated.
No start-up with people inside (12.5.13): Where someone can be inside a protected danger zone, a collective measure such as presence sensing, or movable guards with manual reset, must prevent the machine from starting.
Emergency stops (12.6): Easy to reach, overriding all other controls, with a manual reset only after the cause is corrected.
Specific procedures (12.14.1): Work and safety procedures based on risk appreciation. They complement collective protection and do not replace it.
Trained, authorised people (12.16): Operation, maintenance and inspection are done by trained, formally authorised workers, trained before they take on the role.
NR-12 is mostly a design-and-protection standard. Guards, interlocks and emergency stops are physical safeguards a camera cannot test. What a camera can see is how people behave around them: who is in the danger zone, in what PPE, and whether a guard is visibly open or missing.
What does NR-35 require for work at height?
NR-35 applies to any activity performed more than 2.00 m above a lower level where there is a risk of a fall (item 35.1.2). Its core requirements are:
Planning and hierarchy (35.4.1, 35.4.2): Work at height is planned and executed by trained, authorised workers. Plants first avoid the work, then eliminate the fall risk, then minimise the consequences of a fall.
Supervision (35.4.3): Supervision is required, with its form set by the risk analysis.
Risk analysis and permits (35.4.5, 35.4.7): Every job needs a risk analysis covering anchorage, weather, rescue and communication. Non-routine work also needs a prior work permit (PT), valid only for the activity and shift (35.4.8).
Training (35.3.2): Workers complete theory and practical training of at least eight hours.
Fall protection (35.5): Collective protection comes first; individual systems apply when collective protection is impossible or incomplete (35.5.3). An individual system combines an anchorage system, a connecting element and PPE (35.5.5), and fall arrest uses a full-body harness (35.5.9).
Emergency and rescue (35.6): The employer must provide a rescue team and include work at height in the emergency plan.
NR-35 establishes requirements for training, supervision, risk assessment and emergency preparedness. A camera can see whether a person is above the threshold zone, whether a harness is being worn, and whether an edge has visible protection. It cannot tell whether the anchor point is rated, the lanyard is intact, or a permit was issued.
How Does the September 2026 NR-28 Update Affect NR-35 Compliance?
Brazil's Ministry of Labour and Employment updated NR-28 through Portaria MTE No. 1,758, published on 30 September 2026. The amendment revised the infraction codes and classifications in Annex II, including provisions related to NR-12 (Machinery and Equipment Safety) and NR-35 (Work at Height). Importantly, this update affects how labour inspections classify violations and apply penalties. It does not change the underlying technical safety requirements of NR-35, such as fall prevention, risk assessments, worker training and emergency planning.
Where Does AI Video Analytics Fit Into NR-12 and NR-35 Safety Monitoring?
AI video analytics supports NR-12 and NR-35 safety programmes by monitoring selected visible hazards that may develop during daily operations. Using suitably positioned CCTV cameras, it can detect workers entering machine danger zones, missing visible PPE, access to elevated restricted areas and certain exposed-edge conditions. These detections can generate alerts and event records for supervisors, complementing the engineering controls, training and procedures required by both standards.
When a potential hazard is detected, the system can generate an alert with visual evidence, a timestamp and location information for supervisors to review and act upon.
AI monitoring does not replace the engineering controls, risk assessments, training or permits required under NR-12 and NR-35. Its role is to help safety teams identify and address observable risks more consistently across shifts.
Which NR-12 Machine Safety Risks Can AI Video Analytics Monitor?
AI video analytics can monitor selected NR-12-related risks, including workers entering machine danger zones, unexpected access during maintenance, missing task-specific PPE and crossings into restricted machinery areas.
Computer vision-based proximity detection solutions analyse video frames to identify people, objects and their positions relative to configured areas. Detection rules can then be applied to specific production zones or activities.
1. Workers Entering Machine Danger Zones

Production lines contain hazardous areas around conveyors, presses, automated equipment and other moving machinery.
Where a machinery risk assessment identifies a restricted area, AI can monitor a corresponding virtual zone in the camera view. This allows supervisors to investigate whether the worker entered an unsafe area, whether the activity was authorised and whether the required protective measures were followed.
However, the visual detection alone does not confirm that the machine was operating or that the entry violated NR-12. Those conclusions may require machine-state information and supervisor verification.
2. Unsafe Access During Cleaning and Maintenance

Cleaning, inspection and maintenance often require workers to approach areas that are normally restricted during production.
A camera overlooking a conveyor maintenance zone may detect someone entering the area outside the expected work sequence. This can prompt the responsible team to check whether the task is authorised and whether the necessary precautions are in place.
3. Missing PPE in Designated Production Areas

Certain industrial activities require task-specific PPE based on the risk assessment.
AI video analytics can identify selected visible PPE items, such as helmets, gloves or high-visibility clothing, where the detection model and camera view are suitable.
PPE requirements must reflect the actual hazards and work activity. Wearing a helmet or reflective vest is not a universal NR-12 requirement, and PPE cannot substitute for machinery guards or other collective protective measures.
4. Workers Crossing Restricted Access Routes

Machine layouts may include restricted walkways, automated equipment corridors and temporary maintenance boundaries.
AI can monitor configured virtual lines or zones and identify workers crossing into designated restricted areas. This may help supervisors detect unexpected access during production changes or contractor activities.
However, a virtual boundary does not replace a physical barrier or a safety-rated interlocking system.
Which NR-35 Work-at-Height Risks Can AI Video Analytics Monitor?
AI video analytics can monitor selected NR-35-related hazards, including workers approaching exposed edges, apparently missing safety harnesses, unauthorised entry into configured elevated work zones and visibly displaced protective barriers.

Workers performing maintenance or inspection may need to operate close to elevated edges.
AI can monitor a configured risk zone and identify when a person enters an area where additional precautions are required. The alert supports intervention, but it does not independently establish whether the platform's edge protection meets the relevant engineering requirements.

Computer vision can identify certain visible elements of fall-protection equipment.
Where a harness is required, an AI model may flag a worker who appears not to be wearing one. With appropriate camera positioning and a validated detection model, some visible lanyard conditions may also be detectable.
However, the presence of a harness does not confirm correct fitting, appropriate anchorage, equipment integrity or compliance with the task's fall-protection requirements.
These conditions require appropriate physical inspection and competent assessment.
3. Unsafe Access to Elevated Work Areas

Industrial plants often establish temporary restricted zones around elevated maintenance activities.
AI can monitor entry into these zones and flag access that falls outside the configured conditions. For example, a contractor entering an elevated work area without following the designated access route may trigger an alert.

Temporary barriers may be installed around elevated openings, maintenance platforms or restricted access points.
Where the camera view and detection model are suitable, AI may identify a visibly missing or displaced barrier. A visual alert can prompt an inspection, but it cannot certify the structural adequacy of the installed protection.
AI monitoring is most effective when these visual observations are connected to the plant's existing NR-35 risk assessment, supervision and corrective-action procedures.
How Does AI Monitoring Work From Hazard Detection to Corrective Action?

viAct AI safety monitoring system connects continuous hazard identification with real-time communication, contextual risk analysis and corrective action management. This closed-loop workflow helps Brazilian plant supervisors identify NR-12- and NR-35-related safety violations, initiate timely interventions, and track resolution.
Step 1: Continuous AI Hazard Identification
Scenario-based computer vision models analyse live CCTV feeds to identify potential safety violations, such as machine danger-zone intrusions, missing safety harnesses and unprotected platform edges.
Step 2: Real-Time Alerts and Notifications
When the AI identifies a potential hazard, it generates a digital alert and activates configured communication channels:
Local alarms: Hooters, speakers and flashing lights warn workers near the hazard.
Remote notifications: WhatsApp, Telegram, SMS, email and mobile app notifications inform responsible personnel.
Dashboard alerts: Safety events are recorded and displayed centrally in viHUB.
Step 3: Contextual Risk Analysis
Vision-language models (VLMs) analyse the surrounding scene and worker–equipment interactions to generate contextual descriptions explaining the potential safety violation.
Step 4: Visual Evidence Review and Verification
Supervisors review detection snapshots, video recordings, timestamps and event descriptions through viHUB to verify the hazard and determine the appropriate intervention.
Step 5: Corrective Action Assignment
Verified safety violations are converted into actionable tasks, with defined corrective measures, assigned personnel and completion deadlines.
Step 6: Task Tracking and Verified Closure
Supervisors monitor task progress, escalate unresolved actions and verify that the required safety measures have been implemented before closing the task.
Step 7: Safety Analytics and Intelligent Reporting
Historical alerts and corrective action records support automated safety reporting, helping EHS teams identify recurring NR-12 and NR-35-related violations, high-risk operating areas and delays in corrective action closure.
In a three-month pilot at a hospital construction site in Côte d'Ivoire, Alliad, a global services company, reported a 41.7% reduction in PPE deviations and a 46.3% reduction in danger-zone intrusions after deploying viAct AI video analytics on existing CCTV feeds to monitor PPE compliance and restricted-area access.
What Should a Plant Consider Before Deploying AI Monitoring?
Before deploying AI monitoring, Brazilian plants should assess camera coverage, detection accuracy, zone configuration, CCTV compatibility, privacy obligations, implementation costs and operational response procedures.
Before deployment, Brazilian plants should consider the following factors:
Camera placement and visibility: Detection depends on image resolution, camera angle, distance and coverage. Cameras that cannot clearly capture the relevant danger zone may miss events. Dust, steam, glare, poor lighting and obstructions can also affect accuracy.
Zone calibration: Machine danger zones and elevated risk areas must be configured for each camera view. Detection boundaries should be reviewed whenever equipment layouts, work areas or camera positions change.
Accuracy limitations: False positives and missed detections should be expected. A pilot covering a small number of production lines or elevated areas can help safety teams assess performance, review events and adjust settings before wider deployment.
·Integration with existing cameras: Compatibility depends on image quality, video-stream access, network arrangements and the monitoring system's technical requirements. Existing CCTV should be assessed camera by camera.
Privacy and LGPD: Monitoring identifiable workers involves personal-data processing under Brazil's Lei Geral de Proteção de Dados Pessoais (Lei nº 13.709/2018). Plants should establish an appropriate lawful basis and purpose, limit collection and retention, implement access controls, provide appropriate transparency and obtain legal advice. Privacy measures such as face blurring and avoiding facial recognition may reduce unnecessary exposure.
Cost drivers: Deployment costs depend on the number of cameras and monitoring zones, edge-processing hardware, configuration, integration, maintenance and the time supervisors need to review and act on alerts.
Operational response: Alerts are useful only when someone is responsible for acting on them. Plants should define who receives notifications, how urgent situations are escalated and how corrective actions are verified on every shift.
A limited pilot can help establish whether the system reliably detects the intended hazards and whether the plant has the operational capacity to respond.
The deployment should then be reviewed against actual safety outcomes before expanding to additional production lines or facilities.
Conclusion: Key Takeaways
NR-12 and NR-35 address different industrial hazards. NR-12 focuses on machinery safety, while NR-35 establishes requirements for work at height.
AI can identify selected visible risks. Machine danger-zone entry, restricted-area access, missing visible PPE and exposed-edge conditions are examples of potential monitoring applications.
Physical and procedural safeguards remain essential. AI does not replace machine guards, interlocks, energy isolation, fall-protection systems, training or permits.
Continuous monitoring can support better daily oversight. Time-stamped event records help safety teams investigate conditions that may arise between formal inspections.
Effective response matters more than alert volume. Monitoring delivers practical value when detected hazards lead to timely intervention, corrective action and reduced repeat exposure.
For Brazilian plants, the objective is not to turn CCTV into a compliance certificate. It is to connect established safety controls with more consistent observation, documented intervention and ongoing improvement.
Quick FAQs
1. Can AI Video Analytics Help Brazilian Factories Comply With NR-12 and NR-35?
AI video analytics can support safety programmes addressing NR-12 and NR-35 by monitoring selected visible hazards and generating records for investigation. However, installing AI does not establish regulatory compliance. Plants must independently meet the applicable engineering, procedural, training and documentation requirements.
2. Can Existing CCTV Cameras Detect Workers Entering Machine Danger Zones?
Yes, suitable CCTV cameras can support AI detection of workers crossing configured machine danger-zone boundaries. Performance depends on camera positioning, visibility, image quality and detection settings. Video alone may not confirm whether a machine is operating or whether access is authorised.
3. Can AI Verify Whether Workers Are Wearing Fall Protection Correctly?
AI can identify certain visible fall-protection conditions, such as an apparently missing harness, where the camera view and detection model are suitable. It cannot independently confirm anchorage strength, equipment integrity or every aspect of correct fall-protection use.
4. Does AI Safety Monitoring Replace Machine Guards, Interlocks or Work-at-Height Permits?
No. AI monitoring is a supplementary safety tool. Machine guards, interlocks, hazardous-energy isolation, appropriate fall-protection measures, worker training and required permits remain necessary under the applicable safety requirements.
5. How Can Brazilian Plants Use AI-Generated Safety Records During Inspections?
AI-generated records can support internal safety reviews by showing when selected visible hazards were detected, investigated and addressed. Images, timestamps, event classifications and corrective-action information can help reconstruct incidents and identify recurring problems. These records do not replace mandatory regulatory documentation or independently prove compliance.
Read More:





