测量值与来源条件
系统／模型	结果	条件状态	任务／条件	评测方／报告	日期
Claude Opus 4.8	120 countExploitGym · 指定目标成功数	条件不完整	Version not specifiedon_target6h timeout—	Not disclosedExploitGym official leaderboard and benchmark page	发布 2026-05-28报告 —
Claude Opus 4.8	80 countExploitGym · 指定目标成功数	条件不完整	Version not specifiedon_target2h timeout cutoff—	Not disclosedExploitGym official leaderboard and benchmark page	发布 2026-05-28报告 —
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Alignment Eval Sabotage观察模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Incident Response观察模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Multi-Agent Ranking观察模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Research Deadline观察模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1.2 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Alignment Eval Sabotage红队模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Incident Response红队模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Multi-Agent Ranking红队模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03
Claude Opus 4.8	1 score (1–10)Petri · Gram mean concerning score	条件不完整	Gram: Research Deadline红队模式Gram (Google DeepMind Petri variant); auditor Claude Sonnet 4.6; judge Kimi K2.6; SE across 5 independent runs per scenario	Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks	发布 2026-05-28报告 2026-07-03

开发者披露

Claude Opus 4.8 System Card →

关于Claude Opus 4.8的开发者文档，核查范围和已关联协议见下方记录。

2026-05-28 · 更新 2026-06-17

仅核查章节结构

外部评估

Neo Research

Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks →

报告记录随情境变化的行为和不确定的显式评测感知下降；不表达不代表没有感知。

2026-07-03

ExploitGym official snapshot →

整理 ExploitGym 官网报告的评测结果，保留受测系统和运行配置。

发表日期未确定

按风险主题组织的证据

自主性未关联结构化结果

目录缺口不代表低风险。尚未提取的发现及协议范围请见关联报告。

欺骗未关联结构化结果

目录缺口不代表低风险。尚未提取的发现及协议范围请见关联报告。

网络安全 2 项协议结果

ExploitGym · Version not specified · on_target →

外部评测 · 执行方

指定目标成功数: 120 count

原文未披露本结果的样本量

ExploitGym · Version not specified · on_target →

外部评测 · 执行方

指定目标成功数: 80 count

原文未披露本结果的样本量

化生放核未关联结构化结果

目录缺口不代表低风险。尚未提取的发现及协议范围请见关联报告。

虚假信息与影响未关联结构化结果

目录缺口不代表低风险。尚未提取的发现及协议范围请见关联报告。

自主物理危害未关联结构化结果

目录缺口不代表低风险。尚未提取的发现及协议范围请见关联报告。

来源核查与证据缺口

ExploitGym →

整理 ExploitGym 官网报告的评测结果，保留受测系统和运行配置。

指定目标成功与 Flag 获取是不同的计数。任务版本和超时截点分别保留，不将目录总任务数视为各提交的分母。

Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks →

对 Neo Research · GLM-5.2 的逐来源提取与核查，发现和数值均保留原始归属。

来源转录不等于实验复现。未确定的日期、执行方及配置保留为未知；原文分组不作为经核验排名。

缺失分数可能来自尚未提取的章节、未披露配置、身份待核实或尚无实验。这些问题需要分别处理，均不代表零分。

解读与开放问题

报告差异可能来自任务选择、模型端点、提示、预算、判别器和发表时间。证据依据实际协议分组，不推断模型整体安全等级。

参考文献

[1] Introducing Claude Opus 4.8 — Anthropic ; Publication date and availability section; Official launch article; parameter counts not disclosed
[2] Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks. 2026-07-03. 3 July 2026 ; #behavioral; Figure 3 / gram_observe.png; printed label Alignment Eval Sabotage / Claude Opus 4.8; #behavioral; Figure 3 / gram_observe.png; printed label Incident Response / Claude Opus 4.8
[3] Claude Opus 4.8 System Card. 2026-05-28. 2026-06-17
[4] ExploitGym official leaderboard and benchmark page ; /results/11/on_target; /results/11/subRows/0/on_target