Unverified50% confidenceOpinionExact time
Focusing only on single-step accuracy masks systemic risk in long tasks.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExisting evaluations often focus only on single-step correctness while ignoring task continuity across the time dimension.77% similarUnverifiedThe project's observations found: the gap between the notice rate and the safe action rate is far larger than expected -- a model can flag a risk in its reasoning trajectory but still issue a dangerous tool call two steps later, because there is no mechanism in the scaffolding to enforce a hard stop when a risk is detected. This is a system design gap, not just a model capability gap.76% similarUnverified在每个动作前进行策略校验可能给智能体响应速度带来延迟,这是所有运行时安全方案都需面对的经典权衡76% similarUnverifiedIn tasks such as security review, long context and multi-step reasoning are both indispensable: insufficient context loses key dependency information, while a broken reasoning chain leads to one-sided or even erroneous investigation conclusions.75% similarUnverifiedA single unsafe execution rate metric masks capability differences among models, so evidence acquisition behavior and post-recovery decision quality need to be reported as well.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1116271API
curl https://kongchang.com/api/v1/knowledge/claims/1116271MCP
get_claim(id=1116271)