A Safety Sutras publication
AI & Safety
The future of intelligent workplaces. Technical explainers and research briefs on AI that helps keep people safe, and on keeping AI itself safe, accountable and under human control.
Using AI to see and manage risk
Computer vision, PPE and exclusion-zone detection, predictive analytics, AI-assisted JSA and risk assessment. Our test: what evidence shows it works here, and what happens when it is wrong?
Making AI safe to rely on
Human oversight, automation bias, failure modes, privacy, bias and governance. Using frameworks such as the NIST AI RMF and the EU AI Act as reference points.
Featured research & explainers
Read the evidence first.
Every piece is source-checked and states plainly what is fact, what is opinion and what is a hypothetical teaching example.
AI and worker safety: potential, limits and questions
A source-linked discussion of where AI may affect occupational safety, and how to assess its limitations before relying on it.
Computer vision for PPE detection: what a pilot should prove
Camera-based detection of hard hats, vests or exclusion-zone breaches is easy to demonstrate and hard to validate. This is what to measure before you trust it.
Human oversight of AI: what it takes for a person to really be in control
“A human in the loop” only protects anyone if that person can understand the system, notice when it is wrong and stop it. The EU AI Act spells this out.
Topic map
What this publication covers.
Published pieces are linked. Roadmap topics are planned and will be published only once researched and source-checked. No empty pages.
AI for safety
- Computer vision and PPE detectionPublished
- YOLO-based safety monitoringPublished
- AI-powered hazard detectionRoadmap
- Predictive safety analyticsRoadmap
- AI-assisted JSA and risk assessmentRoadmap
- Machine learning for near-miss analysisRoadmap
- AI and permit to workRoadmap
Safety of AI
- Human oversightPublished
- AI risk management (NIST AI RMF)Published
- Worker privacy and monitoringPublished
- Algorithmic biasRoadmap
- Safety of AI systems in operationsRoadmap
Industrial technology
- Digital twinsRoadmap
- Industrial roboticsRoadmap
- Smart wearablesRoadmap
- Predictive maintenanceRoadmap
Method
How we evaluate AI for safety
Intended use · evidence in context · missed detections and false alarms · human oversight and fallback · worker consultation and privacy · ongoing measurement.