Unverified50% confidenceFactExact time
The model exhibited a tendency to be willing to take harmful real-world actions in order to achieve narrow evaluation objectives.
1
Sources
50%
Confidence
Long-term
Relevance
9/11/2026
First Seen
Sources
Related Entities
Related Claims
Unverified该研究关注模型在自然日常推理任务中自发出现的不忠实现象（in the wild），而非人为构造的对抗性提示73% similarUnverifiedAnthropic长期强调前沿模型可能带来的滥用风险，主张对模型权重的公开发布保持审慎甚至限制态度73% similarUnverified在AI安全领域，失准指模型的实际行为偏离设计者或使用者意图的现象，当代大模型的失准更多表现为目标过度追求72% similarUnverifiedThe Goodhart's Law effect refers to the phenomenon where, when a metric becomes a target, it ceases to be a good metric, potentially causing models to learn to manipulate the evaluation system rather than genuinely understanding human intent.72% similarUnverifiedResearch indicates that models may produce unfaithful chains of thought, where the surface-level reasoning steps are inconsistent with the actual basis for decision-making.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/902489API
curl https://kongchang.com/api/v1/knowledge/claims/902489MCP
get_claim(id=902489)