Unverified50% confidenceFactExact time
OpenAI发布了专项研究报告Sycophancy to Subterfuge，追踪评估模型在多轮施压下维持真实判断的能力
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/17/2026
First Seen
Valid until: 10/15/2026
Sources
Related Claims
UnverifiedOpenAI、Anthropic等机构已开始采用动态基准（Dynamic Benchmarks）策略，定期更新测试集并引入对抗性样本以保持评估效度69% similarUnverifiedOpenAI引入激活探针等监控升级，作为机械可解释性研究的工程化应用来检测潜在有害意图68% similarUnverifiedOpenAI采用宪法AI思路对输出进行价值对齐，并通过红队测试在发布前挖掘模型危险能力边界65% similarUnverifiedOpenAI Preparedness团队的核心目标是在模型具备危险能力的早期阶段发现并评估这些能力62% similarUnverifiedOpenAI引入了激活探针（Activation Probes）监控机制，通过监控模型内部神经网络激活状态检测潜在有害思维模式61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/537780API
curl https://kongchang.com/api/v1/knowledge/claims/537780MCP
get_claim(id=537780)