{"schemaVersion":"nuwa-current-record-1","version":"2026-10-29.1","publication":"published","work":{"id":"metr-frontier-risk-report","title":"Frontier Risk Report (February to March 2026)","aliases":[],"authors":[],"primaryFunction":"evaluation","secondaryFunctions":[],"kind":"technical-report","sourceId":"metr-frontier-risk-report-publication","attributionSourceId":"metr-frontier-risk-report-publication","originalUrl":"https://metr.org/blog/2026-05-19-frontier-risk-report/","publicationDate":"2026-05-19","updatedOn":"2026-10-27","areas":["autonomy","deception"],"scopeNote":{"en":"Relevance is assigned to the research question and stated conditions of the work. A topic tag is not a risk rating.","zh":"风险关联依据研究问题及明确条件建立，主题标签不代表风险评级。"},"protocolIds":[],"contributions":[{"organizationId":"metr","role":"evaluation","sourceId":"metr-frontier-risk-report-publication","locator":"Executive summary and guide to the report; Pilot process"},{"organizationId":"metr","role":"publisher","sourceId":"metr-frontier-risk-report-publication","locator":"Executive summary and guide to the report; Pilot process"}],"summary":{"en":"Pilots an assessment of rogue-deployment risk using evidence about internal agent use at participating AI developers.","zh":"结合参与机构内部智能体使用证据，试行未经授权部署风险评估。"},"note":{"question":null,"method":null,"finding":null,"supports":null,"limits":null,"connections":null},"resources":[{"label":"clearly-malicious transcripts","url":"https://metr.org/blog/2025-10-14-malt-dataset-of-natural-and-prompted-behaviors/"},{"label":"the APPS dataset (Hendrycks et al. 2021)","url":"https://arxiv.org/abs/2105.09938"},{"label":"agents can construct encoded messages","url":"https://arxiv.org/pdf/2507.02737"}],"noteStatus":{"question":"limited","method":"limited","finding":"limited","supports":"limited","limits":"limited","connections":"limited"},"noteAvailability":{"en":"A substantive NUWA research note is not yet available. The repeated draft sections were removed pending review of the original methods, findings and limitations; use the original publication for the research account.","zh":"实质性的 NUWA 研究笔记尚未完成。重复草稿章节已移除，待核查原始方法、发现与限制；研究内容请以原始出版物为准。"},"noteSourceLocator":"No complete section-level method/results/limitations review completed for this note; cached original page and any chart audit retained as discovery evidence.","tags":["risk-monitoring"]},"reading":{"id":"metr-frontier-risk-report","title":"Frontier Risk Report (February to March 2026)","en":{"question":"How can periodic entity-level assessment address risks from developers’ internal AI use?","design":"The February 16–March 16, 2026 pilot involved four developers, access to internal models and raw reasoning, nonpublic organizational information and participant-reviewed disclosure.","findings":["The report separates means, motive and opportunity.","It documents substantial coding capability alongside judgment/reliability limits and exceptions to effective monitoring, rather than treating model benchmark scores as complete risk evidence."],"interpretation":"NUWA interpretation: internal-use risk assessment requires organizational access and controls as well as model capability.","limits":"This reading covers the executive summary, pilot process and selected monitoring findings; it is not a complete audit of the very long report or all confidential evidence.","decisionRelevance":"Model recommendation: cite the dated assessment window and distinguish public observations, developer-provided information and METR inference.","takeaway":"NUWA interpretation: internal-use risk assessment requires organizational access and controls as well as model capability.","connections":[],"scoring":null,"conditions":null,"statistics":null,"usage":"Model recommendation: cite the dated assessment window and distinguish public observations, developer-provided information and METR inference."},"zh":{"question":"周期性机构级评估如何覆盖开发者内部AI使用风险？","design":"2026年2月16日至3月16日试点涉及四家开发者，取得内部模型、原始推理及非公开组织信息，并经参与者审核披露。","findings":["报告分开手段、动机和机会。","强编程能力并存判断可靠性限制，监测有效但有例外，模型分数不是完整风险证据。"],"interpretation":"NUWA 解读：内部使用风险需同时审查组织访问、控制及模型能力。","limits":"此次仅选读执行摘要、试点过程和部分监测结果，未完整审计长报告或保密证据。","decisionRelevance":"模型建议：引用评估窗口，并区分公开观察、开发者信息及METR推断。","takeaway":"NUWA 解读：内部使用风险需同时审查组织访问、控制及模型能力。","connections":[],"scoring":null,"conditions":null,"statistics":null,"usage":"模型建议：引用评估窗口，并区分公开观察、开发者信息及METR推断。"},"assertionPolicy":{"question":"source-research-question","design":"source-finding","findings":"source-finding","interpretation":"nuwa-interpretation","limits":"source-limit-or-explicit-editorial-scope-limit","connections":"source-supported-relationship-only","decisionRelevance":"model-recommendation-not-source-conclusion"},"sourceRefs":[{"sourceId":"metr-frontier-risk-report-publication","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","locator":"Executive summary and guide to the report","version":"Selected original resource https://metr.org/blog/2026-05-19-frontier-risk-report/; sha256:b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a; archived source capture; exact capture time not independently established","artifactSha256":"b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a"},{"sourceId":"metr-frontier-risk-report-publication","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","locator":"Pilot process","version":"Selected original resource https://metr.org/blog/2026-05-19-frontier-risk-report/; sha256:b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a; archived source capture; exact capture time not independently established","artifactSha256":"b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a"},{"sourceId":"metr-frontier-risk-report-publication","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","locator":"Opportunity: Monitors caught many harmful actions but there were exceptions and workarounds","version":"Selected original resource https://metr.org/blog/2026-05-19-frontier-risk-report/; sha256:b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a; archived source capture; exact capture time not independently established","artifactSha256":"b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a"},{"sourceId":"metr-frontier-risk-report-publication","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","locator":"No-reasoning and visible reasoning results","version":"Selected original resource https://metr.org/blog/2026-05-19-frontier-risk-report/; sha256:b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a; archived source capture; exact capture time not independently established","artifactSha256":"b0ebbde94fc97264cfac3378eb3794a8caba87ff4a51d35f854a71a565696d8a"}],"publication":{"state":"published","publishedOn":"2026-10-29","authorizationReference":"user-authorized-global-index-publication-2026-10-29","reviewedCandidateId":"local-6efe31bc4a0c0f37"},"review":{"state":"published-note","sourceScope":"full-text","availableScope":null,"readingCoverage":"selected-source-sections; not a complete section-by-section paper audit","completePaperAudit":false,"independentEvidenceReview":"pending-human-review","bilingualEditorialReview":"pending-human-review","authorship":"model-assisted-research-note","preparedFingerprint":"4280e056f9063fef0e88c3bd0499a0824b66733ca2117d1b7ec7a2143cdc4fcf"}},"sources":[{"id":"metr-frontier-risk-report-publication","title":"Frontier Risk Report (February to March 2026)","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","version":"Source capture 2026-10-25","publicationDate":"2026-05-19","datePrecision":"day","type":"research-publication","verifiedOn":"2026-10-27","documentId":"https://metr.org/blog/2026-05-19-frontier-risk-report","authors":[]}]}