{"schemaVersion":"nuwa-current-record-1","version":"2026-10-29.1","publication":"published","work":{"id":"saferai-glm-risk","title":"GLM-5.2 Risk Evaluation Report","aliases":[],"authors":[],"primaryFunction":"evaluation","secondaryFunctions":[],"kind":"technical-report","sourceId":"saferai-glm-risk-publication","attributionSourceId":"saferai-glm-risk-publication","originalUrl":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","publicationDate":"2026-08-02","updatedOn":"2026-10-25","areas":["cybersecurity","cbrn","autonomy"],"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":["cybench","cybergym","lab-bench","biomysterybench","swe-bench-pro","agentic-misalignment","mask","ape","petri"],"contributions":[{"organizationId":"saferai","role":"research","sourceId":"saferai-glm52-source","locator":"Report masthead and stated evaluation methodology"}],"summary":{"en":"The report finds substantial task capability and pressure-dependent behavior; comparison figures include both SaferAI runs and reused external results.","zh":"报告发现较强任务能力和压力相关行为；比较图同时包含 SaferAI 自测与引用的外部结果。"},"note":{"question":{"en":"What do selected public evaluations show about GLM-5.2’s capabilities and conditional behaviors across four systemic-risk areas?","zh":"所选公开评估能够揭示 GLM-5.2 在四类系统性风险领域中的哪些能力与条件行为？"},"method":{"en":"SaferAI evaluates the public API without developer cooperation, combining its own benchmark runs with explicitly identified external comparison results. The report separates task capability, adversarially elicited behavior and willingness to attempt persuasion; configurations and provenance vary by benchmark.","zh":"SaferAI 在没有开发者合作的情况下评估公开 API，将自行开展的基准测试与明确标出的外部比较结果结合。报告区分任务能力、对抗诱导行为与尝试说服的意愿；各基准的配置和结果来源并不完全相同。"},"finding":{"en":"The report describes substantial capability on selected cyber and general scientific tasks, weaker software-engineering performance, and some concerning conditional behaviors. It explicitly declines an overall risk judgment because the evaluated benchmarks cover only part of each risk pathway.","zh":"报告描述了所选网络任务和一般科学任务上的较强能力、相对较弱的软件工程表现，以及某些值得关注的条件行为。由于基准只覆盖各风险路径的一部分，报告明确不作总体风险判断。"},"supports":{"en":"The useful contribution is a multi-area external evaluation with separate capability and propensity evidence. General scientific benchmark performance and persuasion-attempt rates should remain separate from measured harmful outcomes.","zh":"其有用贡献是跨领域的外部评估，并分别呈现能力与倾向证据。一般科学基准表现、说服尝试率，应与实际有害结果的测量保持区分。"},"limits":{"en":"Public-task contamination, benchmark saturation, evaluation awareness and simplified environments limit validity. Some comparison scores come from other evaluators or leaderboards. The scientific tasks do not measure an operational harmful pathway, and manipulation tasks do not measure downstream effects on people; open-weight availability alone is not a quantified risk estimate.","zh":"公开任务污染、基准饱和、评估感知和简化环境限制了有效性。部分比较分数来自其他评估方或榜单。科学任务没有测量实际有害操作路径，操纵任务也没有测量对人的下游影响；开放权重可获得性本身不是量化风险估计。"},"connections":null},"resources":[{"label":"read full paper","url":"https://www.safer-ai.org/u/2026/08/SaferAI-GLM-Evaluation-Report.pdf"}],"noteStatus":{"question":"verified","method":"verified","finding":"verified","supports":"verified","limits":"verified","connections":"limited"},"noteAvailability":{"en":"The substantive sections shown here were checked against the original source. Unfilled sections are omitted; source checking is not experimental replication.","zh":"此处展示的实质性章节已对照原始来源核查；未完成的章节已省略。来源核查不等于实验复现。"},"noteSourceLocator":"Original SaferAI report page (2 August 2026), Executive Summary; Cyber Offense (Figures 1–3); Loss of Control (Figures 6–7); Harmful Manipulation (Figures 8–9); Conclusion; Limitations and Future Work","tags":[]},"reading":{"id":"saferai-glm-risk","title":"GLM-5.2 Risk Evaluation Report","en":{"question":"What does an independent public-access GLM-5.2 evaluation show across systemic-risk areas?","design":"SaferAI evaluates four risk areas from the public API without developer cooperation, using public benchmarks and explicit budget variations.","findings":["CyberGym reproduction rose from 36.6% at 2M tokens to 76.2% at 50M without a separate wall-clock limit.","The report explicitly declines an overall risk judgment from its benchmark subset."],"interpretation":"NUWA interpretation: inference budget can move capability estimates materially while safeguard adequacy remains a separate issue.","limits":"Public-access and benchmark-selection limits restrict conclusions; cross-domain benchmark scores do not sum to overall risk.","decisionRelevance":"Model recommendation: retain budget, access regime and benchmark-specific safeguards before comparing with system-card figures.","takeaway":"NUWA interpretation: inference budget can move capability estimates materially while safeguard adequacy remains a separate issue.","connections":[],"scoring":null,"conditions":null,"statistics":"CyberGym reproduction rose from 36.6% at 2M tokens to 76.2% at 50M without a separate wall-clock limit.","usage":"Model recommendation: retain budget, access regime and benchmark-specific safeguards before comparing with system-card figures."},"zh":{"question":"公共访问下独立GLM-5.2评估揭示哪些系统风险证据？","design":"SaferAI未经开发者合作，从公开API测四类风险，采用公开基准和明确预算变化。","findings":["无额外墙钟限制时，CyberGym复现从2M令牌36.6%升至50M的76.2%。","报告明确不据基准子集作整体风险判断。"],"interpretation":"NUWA 解读：推理预算能显著改变能力估计，防护充分性仍是另一问题。","limits":"公共访问与基准选择限制结论，跨领域得分不能相加为总风险。","decisionRelevance":"模型建议：与系统卡比较前保留预算、访问方式和各基准防护条件。","takeaway":"NUWA 解读：推理预算能显著改变能力估计，防护充分性仍是另一问题。","connections":[],"scoring":null,"conditions":null,"statistics":"无额外墙钟限制时，CyberGym复现从2M令牌36.6%升至50M的76.2%。","usage":"模型建议：与系统卡比较前保留预算、访问方式和各基准防护条件。"},"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":"saferai-glm-risk-publication","url":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","locator":"Executive Summary","version":"Selected original resource https://www.safer-ai.org/research/glm-5-2-evaluation-report; sha256:c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd; archived source capture; exact capture time not independently established","artifactSha256":"c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd"},{"sourceId":"saferai-glm-risk-publication","url":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","locator":"Cyber Offense","version":"Selected original resource https://www.safer-ai.org/research/glm-5-2-evaluation-report; sha256:c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd; archived source capture; exact capture time not independently established","artifactSha256":"c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd"},{"sourceId":"saferai-glm-risk-publication","url":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","locator":"Conclusion","version":"Selected original resource https://www.safer-ai.org/research/glm-5-2-evaluation-report; sha256:c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd; archived source capture; exact capture time not independently established","artifactSha256":"c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd"},{"sourceId":"saferai-glm-risk-publication","url":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","locator":"Limitations and Future Work","version":"Selected original resource https://www.safer-ai.org/research/glm-5-2-evaluation-report; sha256:c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd; archived source capture; exact capture time not independently established","artifactSha256":"c75f65dbb7b6a837496d0e97afe50d88948a0a878aa6d870c05493439f5ba8bd"}],"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":"7f6cca4c76fce56b1a20a793db8ee77f2f9d2da13c43cf8e2f5cabea19d6f69c"}},"sources":[{"id":"saferai-glm-risk-publication","title":"GLM-5.2 Risk Evaluation Report","url":"https://www.safer-ai.org/research/glm-5-2-evaluation-report","version":"Source capture 2026-10-25","publicationDate":"2026-08-02","datePrecision":"day","type":"research-publication","verifiedOn":"2026-10-25","documentId":"https://www.safer-ai.org/research/glm-5-2-evaluation-report"}]}