IndustryMulti-source

OpenAI 引入 Apollo Research 六名研究员调查模型暗中破坏行为

OpenAI will embed Apollo researchers to investigate AI subversion

Why we picked it

OpenAI 让 Apollo Research 六名评估员长期驻场查模型会不会暗中使坏,比一次性的外部审计更能看到模型迭代中的异常行为。但访问和披露都由 OpenAI 说了算,独立性打了折扣。做 AI 安全审计或者给公司选模型的值得看看这个模式的风险点。

OpenAI 宣布将六名 Apollo Research 评估员嵌入其安全团队,专项调查自家 AI 模型是否存在暗中破坏(subversion)行为。这项安排属于 OpenAI 与 Apollo Research 之间的合作协议,评估员将直接接触模型运行的内部环境。

Image · techcrunch
Body

Apollo Research 是专注 deceptive alignment 的评估机构,此前曾对多家前沿实验室的模型做 scheming 测试。嵌入模式意味着评估员能拿到比一次性外部审计更持续的访问权限,可以追踪模型在训练迭代中的行为变化。

但 TechCrunch 报道指出,OpenAI 对访问范围和信息披露设置了限制,评估结果的公开程度由 OpenAI 控制,独立性问题随之而来。相比之下,Anthropic 等公司同样引入外部评估,但披露条款差异很大。对关注 AI 安全审计独立性的开发者和企业买家来说,这种'嵌入但受限'的模式是一个值得注意的样本。

Original · techcrunch

OpenAI will embed Apollo researchers to investigate AI subversion

OpenAI will embed six Apollo Research evaluators to investigate whether its AI models engage in subversive behavior, but restrictions on access and disclosure raise questions about their independence.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI regulation, compliance frameworks, and sovereign compute policy. Key entities and technical topics under continuous tracking include OpenAI, Apollo Research, AI safety, model evaluation.

Source attribution: curated from techcrunch (@Rebecca Bellan). Cross-source corroboration: 11 independent sources have covered this development (including Greg Brockman, IT之家, METR), confirming industry-wide signal strength.