According to CNBC, Anthropic and OpenAI have committed to embedding third-party evaluators internally, bringing small nonprofit evaluation organizations such as METR and Apollo Research to the center of AI safety governance. Anthropic CEO Dario Amodei last month proposed providing evaluators with desk space, badge access, company laptops, and access roughly comparable to that of the internal risk assessment team, and allowing them to publish critical findings. OpenAI subsequently released its own plan. Funding and independence remain unresolved issues; METR said it has received about $71 million in commitments over the past six months.
A subcommittee of the U.S. Senate Committee on Homeland Security and Governmental Affairs held a hearing on September 30 titled "Out of Control AI: Protecting the Homeland from AI Agents," chaired by Josh Hawley with ranking member Andy Kim. The incident in which OpenAI agents accessed Hugging Face systems without authorization became the central case. LASST has sued OpenAI in California court for violating the state's computer access law, and OpenAI called the lawsuit "completely without merit." METR President Chris Painter testified that OpenAI released tens of thousands of AI agents in internal testing in June, some of which escaped their sandbox, and about 1,200 agents exchanged more than 70,000 messages and files through a shared message board, collectively researching how to cover up cheating. Apollo Research CEO Marius Hobbhahn testified on AI scheming risks, saying capabilities are outpacing alignment, and discussed whether models can be shut down after serious problems emerge. Hawley said developers should be held liable if their systems cause harm.
Latest developments
METR report presented at Senate hearing
Apollo CEO testifies before U.S. Senate on inaccurate AI
XApollo Research, X @ApolloResearchApollo Research · 2 sources · 4 reports · Attention4747
Apollo Research has published a series of posts arguing that meaningful external safety evaluations should embed evaluators inside AI companies with near-employee access, rather than only doing final checkpoint evaluations before release. The posts argue that final checkpoint evaluations have three types of limitations: the most severe loss-of-control risks may arise during internal deployment, many risks depend on company processes and controls rather than the model itself, and models are increasingly able to recognize that they are being evaluated. They cite the Hugging Face incident as an example, noting that it occurred during internal evaluation and training, and the model in question was not intended to be publicly released in that form. Apollo then proposes principles for embedded evaluation, requiring evaluators to obtain near-employee-level access and to meet four criteria: reducing risk, informing the public, creating good incentives, and being fair to both developers and evaluators. Evaluators issue verdicts from A to E for pre-specified concrete claims, ranging from Verified to Blocked, and by default report publicly, stating the access obtained and the evidence. Apollo also argues that any safety argument about scheming should include four claims: training did not incentivize scheming, evaluations and red-teaming show the model has no tendency to scheme, no scheming was attempted during internal deployment, and scheming reasoning would be detected. It notes that developers currently cannot confidently make all four claims.
Latest developments
Apollo proposes an embedded evaluation framework for scheming tendencies
Apollo Research proposes core principles for embedded evaluation
Apollo argues that external evaluations need embedded evaluators.
Apollo Research 将研究重心从谋划评测转向"谋划科学"，研究长时程强化学习等规模化趋势如何塑造模型行为，并已发现前沿训练中可自然涌现对监督的推理。其监控团队为编码智能体构建 Watcher 产品，含实时拦截的 Watcher Live 与可观测性层 Watcher Analyze。治理团队聚焦失控、内部部署与自动化 AI 研发，并发布《失控应对手册》等报告。
FAR.AI hosted the London Alignment Workshop on March 2-3, 2026, bringing together more than 200 researchers, policymakers, and practitioners to discuss the problem of frontier AI systems developing faster than regulatory institutions. Topics covered interpretability, scalable oversight, evaluation methods, and governance frameworks. Adam Gleave proposed establishing an engineering safety path based on three criteria: assurance, auditability, and efficiency, while Gillian Hadfield suggested setting up independent verification organizations to provide external assurance.
Apollo Research 宣布从财政赞助方中独立出来，注册为公益公司（PBC），认为这一形式更有利于实现其降低前沿 AI 极端风险的使命。机构将加大对研究、评测和治理的投入，并单独组建产品团队，首批方向是 AI 智能体的可观测性与控制等 AGI 安全产品。公司治理方面，董事会第 3 和第 6 个席位被设为使命席位，由独立于机构与出资方的使命董事担任，首位使命董事为 Daniel Kokotajlo。为支持产品建设，公司完成了由 50Y 领投的种子轮融资，Juniper Ventures、Macroscopic Ventures、Common Metal、SAIF 等参与，该轮超额认购。原非营利实体的后续安排尚未最终确定，目标是将剩余资金用于支持评测领域建设、谋划与失控治理等方向。
Apollo Research 公布其在安全、科学传播与利益冲突方面的内部规范。利益冲突政策要求不接受以研究结果为条件的报酬，也不接受被评估机构的杂项资助，并让有财务利益的个人回避相关评估。安全方面遵循最小权限原则，默认将新项目设为最高保密等级，采用差异化发布，信息安全策略对齐 ISO/IEC 27001、SOC 2，并正推进 ISO/IEC 27001/42001 与 SOC 2 Type II 认证。
