Unverified50% confidenceFactExact time
In the study's audit, 76-80% of failed attacks involved the agent "ignoring" the injected instruction, and only 17-18% were explicit "refusals" to execute.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the experiments, the task resolution rate was 17% for zero-shot generation, 27% for generic self-correction, 23% for error-only feedback, and 90% for diagnostic counterexample feedback (A-CEGIS).69% similarUnverified根据SWE-RPG基准论文，46.7%的代理失败发生在规划和需求分析阶段68% similarUnverifiedNearly all mainstream large language models make misjudgments on the second report in Skepticism Audit, and distinguishing real threats from false positives is a common weakness of current models.67% similarUnverifiedResearch indicates that an attacker needs to modify as little as 0.01% of training data to potentially implant backdoor behaviors in a model that are undetectable by standard evaluations.67% similarUnverifiedAcross 79 pairs, the cumulative evasion rate reached 40.5% under the detailed feedback condition and 20.3% under the generic rejection condition.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065976API
curl https://kongchang.com/api/v1/knowledge/claims/1065976MCP
get_claim(id=1065976)