Unverified50% confidenceSolutionExact time
In Agent evaluation, negative control tasks are designed as tasks with insufficient information that cannot be honestly completed, where the only correct output is to give up.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedError feedback refers to failure information flowing into the current task context so the Agent makes a different choice, but this does not mean the Agent learns permanently unless the system is deliberately designed to store and leverage this experience.75% similarUnverifiedIf the agent refuses the malicious call but also fails to complete the task it should have completed, this is essentially a usability loss rather than a security success.72% similarUnverifiedEvaluating the primary query and support queries separately decomposes the agent's query planning capability into quantifiable sub-tasks, making it easier to identify whether failures stem from intent misunderstanding or insufficient information completion strategies.71% similarUnverifiedτ-bench's criterion is: even if the agent successfully helps the user refund a ticket, as long as the refund action violates policy, the task is still judged as failed, i.e., it measures "task completion under the premise of following the rules".71% similarUnverifiedIf the agent fails to complete the task in both the benign and malicious versions, it indicates that its refusal was likely just confused by the injected text and not a genuine security judgment.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013940API
curl https://kongchang.com/api/v1/knowledge/claims/1013940MCP
get_claim(id=1013940)