Identifying the Risks of LM Agents with an LM-Emulated Sandbox
The framework uncovers plausible high-consequence failures without implementing every real tool. Human checks also reveal simulator errors and missed unsafe actions.
Verify byline in the original ↗