Unverified50% confidenceTradeoffExact time
Offline evaluation sacrifices some test coverage of real-world scenarios but in return achieves effective risk control.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/11/2026
First Seen
Valid until: 1/9/2027
Sources
Related Entities
Related Claims
UnverifiedThe risk of false positives—incorrectly classifying normal agent behavior as out-of-bounds and quarantining it—will directly affect the platform's practical utility.76% similarUnverifiedVersion drift is dangerous in evaluation scenarios because the validity of historical measurement results implicitly decays after changes to model weights, inference frameworks, or test environments74% similarUnverifiedFocusing on a single narrow scenario reduces the risk exposure of a pilot, making the impact of errors predictable and reversible.73% similarUnverified底层模型更换后，即便微小的输出分布变化也可能导致此前校准的安全护栏过度拦截或防护失效，需要重新标注和调试72% similarUnverifiedThere is an inherent tension in dangerous capability assessments between the coverage, update frequency, and confidentiality of test sets: overly public test sets become incorporated into training data and lose effectiveness, while overly confidential tests lack peer-review credibility.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/999027API
curl https://kongchang.com/api/v1/knowledge/claims/999027MCP
get_claim(id=999027)