Kshitij Mishra
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
PROBE: Learning to Audit Policy Compliance in Tool-Using LLM Agents
According to the abstract excerpt, the authors argue that existing tests miss whether deployed agents follow rules on pricing, privacy or confirmation during valid tool use. They propose PROBE, which trains auditor models to talk with agents as realistic users and probe for violations.