Dictionary entry Trending
agentic safety
also agent safety or agentic AI safety
Definition of agentic safety
-
: the field of making AI agents safe to deploy when they can plan, use tools, access data, and take multi-step actions, including controls for permissions, monitoring, containment, verification, and human intervention
- The company treated agentic safety as a runtime problem because the assistant could send messages and modify production records.
- Agentic safety testing included prompt injection, failed tool calls, unauthorized actions, and attempts to bypass the shutdown control.
Origin & history
Safety work on autonomous agents predates large language models, but agentic safety became a recognizable label as model-based agents gained tools, memory, and permission to act in software and physical environments. By 2026, research reviews, public funding programs, and major technology companies were using the term for safety and security work focused on agent behavior across complete trajectories rather than isolated model outputs.
Editorial references
Sources & further reading
- Agentic AI Safety: A Structured Review of Open Problems and Their Regulatory Anchoring↗ Peer Reviewed Review · AI · 2026
- Agent Safety Should Be a Runtime Contract↗ Research Paper · arXiv · 2026-08-11
- NVIDIA Commits $250 Million to Advance UK Agentic Safety and Security Research Over the Next Five Years↗ Official Announcement · NVIDIA · 2026-10-13
Test yourself
Which of these is the meaning of agentic safety?
Cite this entry
"Agentic safety." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/agentic-safety/. Accessed 15 Oct. 2026.