Skip to main content
All glossary terms

AI governance

AI Safety

AI safety is the technical and organizational work of reducing the likelihood and impact of harmful or unreliable behavior from artificial intelligence systems. It can include testing, access boundaries, monitoring, human review, and response planning, selected for a system’s capabilities and use context.

Also known as artificial intelligence safety.

Why it matters in practice

Safety is broader than an individual model test because system behavior also depends on data, tools, users, and deployment choices. Teams should state which hazards they address and how evidence supports the control. Calling a system safe without a defined scope, assumptions, and ongoing review can obscure unresolved risk.

Example

Before connecting an assistant to a live account tool, its team tests misuse cases and restricts the tool to read-only actions until review is complete.

Educational note

This definition is general information, not legal advice. It may omit facts that matter to a specific situation and does not establish or guarantee compliance.