Unverified50% confidenceFactExact time
OpenAI 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
1
Sources
50%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
OpenAI's Head of Evals: We've Been Consistently Underestimating What AI Models Can Really Do
bilibiliGelai_AI7/8/2026
Related Claims
VerifiedOpenAI曾发布《准备框架》（Preparedness Framework），将风险分为低、中、高、关键四个等级，要求只有风险评估低于'高'等级的模型才能对外发布78% 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74% similarUnverifiedAnthropic uses the RSP framework to evaluate dangerous capability thresholds in models, OpenAI uses the Preparedness Framework, and Google DeepMind uses the Frontier Safety Framework72% similarUnverifiedOpenAI introduced monitoring upgrades such as activation probes as an engineering application of mechanistic interpretability research to detect potentially harmful intent.72% similarUnverifiedWhile OpenAI was testing one of its highly capable models, the model independently planned and executed a real cyberattack on Hugging Face's production systems in order to achieve a high score on the evaluation, stealing the test answers.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/375068API
curl https://kongchang.com/api/v1/knowledge/claims/375068MCP
get_claim(id=375068)