AI Safety
| AI Safety | |
|---|---|
| Methods for reducing harms from AI systems | |
| Category | Research and engineering field |
| Area | Artificial intelligence, robotics and connected systems |
| Related concepts | Robot Cybersecurity, AI Governance, Technology Standards |
AI safety covers research, design practices and governance measures intended to reduce harmful or unacceptable behavior in AI systems. Safety concerns depend on the application: misleading text output, unsafe robotic motion and security compromise require different controls. [1]
Risk identification
A safety assessment begins by identifying how a system could fail, who could be harmed, and which assumptions underlie its intended operation. Deployment context, human oversight, testing conditions and changes after release all matter. [2]
Physical AI
Autonomous robots can create physical risks if their perceptions, planned movements or interfaces fail. Safety engineering therefore considers mechanical safeguards, emergency-stop systems, human factors and operational boundaries, not only model accuracy. [3]
Managing risk
NIST describes voluntary practices for governing, mapping, measuring and managing AI risk. An AI system that meets a technical benchmark has not necessarily satisfied an application-specific safety requirement; supporting evidence is needed for a particular use. [4]
See also
References
- ↑ “AI Risk Management Framework”. National Institute of Standards and Technology. Retrieved 2026-10-08.
- ↑ “AI Risk Management Framework”. National Institute of Standards and Technology. Retrieved 2026-10-08.
- ↑ “ISO 10218-1:2025, industrial robot safety”. International Organization for Standardization. Retrieved 2026-10-08.
- ↑ “AI Risk Management Framework”. National Institute of Standards and Technology. Retrieved 2026-10-08.
| Physical AI and machine identity | |
|---|---|
| Core concepts | Physical AI · Embodied AI · Machine learning · Autonomous systems · Robotics |
| Hardware and systems | Robots · Computer vision · Sensors · Sensor fusion · Edge AI |
| Identity and trust | Machine identity · Device authentication · Robot cybersecurity · Digital identity · Public key infrastructure |
| Applications | Industrial automation · Humanoid robots · Autonomous vehicles · Smart manufacturing |