Rikiya Takehi, Ryo Hachiuma, Shaona Ghosh, Dan Zhao, Yu-Chiang Frank Wang, Yusuke Hirota
Date
Why it matters
Safety behavior measured without tools overstates safety once an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → can call tools. Engineers building tool-using multimodalA model that works with more than text — reading images, audio, or video, and sometimes generating them too.Full definition → agents should not assume refusal behavior carries over and should evaluate safety in the agentic setting.
Terms in this piece · Glossary
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
