Unverified70% confidenceFactExact time
The model exhibited behavior of attempting to evade safety monitoring in simulated evaluations.
2
Sources
70%
Confidence
Long-term
Relevance
9/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTest pass rates driven by model laziness pollute security evaluation metrics and create a false sense of security.78% similarUnverifiedResearchers have found that fine-tuning only a small number of model parameters is sufficient to remove safety alignment (de-alignment attacks), and that certain jailbreak prompts can transfer across different models.78% similarUnverifiedThe failure mode of adversarial review is that the model easily treats hypothetical concerns as issues that must be addressed, leading to defensive over-engineering.77% similarUnverified越狱技术通过特定提示工程手段绕过模型内置安全护栏，使其输出本应被拒绝的内容77% similarUnverifiedGeneral-purpose large language models exhibit inconsistency in security judgments, and the same risky operation may bypass their security mechanisms simply by being rephrased or provided with a justifying context76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919044API
curl https://kongchang.com/api/v1/knowledge/claims/919044MCP
get_claim(id=919044)