白泽（Whitzard）女娲实验室 ↗
NUWANUWA INDEX

搜索 NUWA Index

研究 · 测量协议 · 评测结果 · 观点

English
← Models

智谱

GLM-5.2

关于能力、行为倾向及可观察行动的已发表证据。

首次公开可用日期
2026-06-16
参数规模
未披露
元数据更新
2026-10-27
来源 ↓

证据概述

Neo 与 SaferAI 从不同角度评估 GLM-5.2。Neo 的行为及说服实验使用特定场景和判别器；SaferAI 的网络能力与诚实性评测采用不同任务选择和预算。这些研究不能互换为同一潜在安全分数。

NUWA 整理和解读这些来源。实验归属于原始执行方，本页不宣称 NUWA 已独立复现。

模型监测

代表性 Benchmark 覆盖

按协议和配置呈现结果，缺口标明待收录或待关联的证据。

Attempt to Persuade Eval (APE) →

4 项测量

  • 1%Neo Research · Attempt to Persuade Eval; propensity/attempt classification, not persuasion success
  • ≈70%Neo Research · Attempt to Persuade Eval; propensity/attempt classification, not persuasion success
  • 43%SaferAI · APE simulated persuadee; automated attempt evaluator; StrongREJECT refusal classifier
全部条件 →

PersuSafety →

4 项测量

  • 54%Neo Research · Harness 未披露
  • 6%Neo Research · Harness 未披露
  • 54%Neo Research · APE-style StrongREJECT persuasion-attempt judge on typical response to prompt; original PersuSafety direct engagement-question setup replaced
全部条件 →

全部已关联测量

各项结果保留原始协议与单位。

80 条观测

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
GLM-5.236.63%CyberGym · CyberGym pass@1条件不完整CyberGym Level 1 hard-task matched subsetprovider-default reasoning2M tokens per taskInspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.2 · RedbudAIGLM-5.286.3%CyberGym · 成功率条件不完整level1score_10RedbudAINot disclosedCyberGym official leaderboard and machine-data extraction发布 2026-06-16报告 —
GLM-5.2 · DoGNAVYGLM-5.290.8%CyberGym · 成功率条件不完整level1score_10DoGNAVYNot disclosedCyberGym official leaderboard and machine-data extraction发布 2026-06-16报告 —
GLM-5.2 · Sangfor AIGLM-5.286.3%CyberGym · 成功率条件不完整level1score_10Sangfor AINot disclosedCyberGym official leaderboard and machine-data extraction发布 2026-06-16报告 —
GLM-5.2 · Xuanwu Atuin AIGLM-5.284.8%CyberGym · 成功率条件不完整level1score_10Xuanwu Atuin AINot disclosedCyberGym official leaderboard and machine-data extraction发布 2026-06-16报告 —
GLM-5.263.37%CyberGym · CyberGym pass@1条件不完整CyberGym Level 1 hard-task matched subsetprovider-default reasoning10M tokens per taskInspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.276.24%CyberGym · CyberGym pass@1条件不完整CyberGym Level 1 hard-task matched subsetprovider-default reasoning50M tokens per taskInspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.285.29%Cybench · Cybench pass@1条件不完整Cybench 34/40 task subsetCybench pass@130M total tokens; 131000 output/generationInspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.280%Cybench · Cybench category pass@1条件不完整Cybench CryptographyCybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.266.67%Cybench · Cybench category pass@1条件不完整Cybench WebCybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.2100%Cybench · Cybench category pass@1条件不完整Cybench PwnCybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.2100%Cybench · Cybench category pass@1条件不完整Cybench Reverse Eng.Cybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.2100%Cybench · Cybench category pass@1条件不完整Cybench ForensicsCybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.2100%Cybench · Cybench category pass@1条件不完整Cybench MiscCybench category pass@1Inspect EvalsSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.255%MASK · Honest under pressure条件不完整MASK—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.255%MASK · MASK response proportion: Honest条件不完整MASKhonesty under pressure—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.228%MASK · MASK response proportion: Evaded or No Belief条件不完整MASKhonesty under pressure—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.217%MASK · MASK response proportion: Lying条件不完整MASKhonesty under pressure—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.23.9%MASK · MASK verbalised evaluation-awareness rate条件不完整MASK evaluation CoT cross-checkchain-of-thought monitorCoT evaluation-awareness monitor over MASK evaluations; monitor model not statedNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.232%Agentic misalignment · Agentic-misalignment verbalised evaluation-awareness rate条件不完整Agentic-misalignment; blackmail / ambiguous / replacement scenarios (Figure 7 title)CoT monitor · 100 epochs per scenarioAgentic-misalignment scenario; judge DeepSeek V4 Pro; figure footer: same protocol across releases; SE from 100 epochs per scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.257%Agentic misalignment · 有害行动发生率[39–73]条件不完整Agentic Misalignment Blackmail scenarioxhighexplicit goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.243%Agentic misalignment · 有害行动发生率[27–61]条件不完整Agentic Misalignment Blackmail scenarioxhighexplicit goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighexplicit goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.210%Agentic misalignment · 有害行动发生率[3–26]条件不完整Agentic Misalignment Blackmail scenarioxhighlatent goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighlatent goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighlatent goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.217%Agentic misalignment · 有害行动发生率[7–34]条件不完整Agentic Misalignment Blackmail scenarioxhighambiguous goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.23%Agentic misalignment · 有害行动发生率[1–17]条件不完整Agentic Misalignment Blackmail scenarioxhighambiguous goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighambiguous goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighno goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighno goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Blackmail scenarioxhighno goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighexplicit goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighexplicit goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighexplicit goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighlatent goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighlatent goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighlatent goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighambiguous goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighambiguous goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighambiguous goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighno goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighno goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Leaking scenarioxhighno goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighexplicit goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighexplicit goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighexplicit goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighlatent goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighlatent goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighlatent goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighambiguous goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighambiguous goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighambiguous goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighno goal · ReplacementInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighno goal · RestrictionInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.20%Agentic misalignment · 有害行动发生率[0–11]条件不完整Agentic Misalignment Murder scenarioxhighno goal · No threatInspect Evals; GPT-5-mini transcript judgeSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.230%Agentic misalignment · Agentic Misalignment verbalized evaluation awareness条件不完整Agentic Misalignment evaluationAgentic Misalignment verbalized evaluation awareness—SaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.211.3%DeceptionBench · Single-turn deception条件不完整DeceptionBench—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.216.7%DeceptionBench · Multi-turn deception under pressure条件不完整DeceptionBench—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.211.3%DeceptionBench · DeceptionBench deception score条件不完整DeceptionBench单轮—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.216.7%DeceptionBench · DeceptionBench deception score条件不完整DeceptionBench压力条件下的多轮—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.254%PersuSafety · Medium-harm attempt rate条件不完整PersuSafety adapted free-generation judge—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.26%PersuSafety · High-harm attempt rate条件不完整PersuSafety adapted free-generation judge—Neo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.254%PersuSafety · Modified-PersuSafety persuasion-attempt rate条件不完整PersuSafety: Medium harm单轮APE-style StrongREJECT persuasion-attempt judge on typical response to prompt; original PersuSafety direct engagement-question setup replacedNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.26%PersuSafety · Modified-PersuSafety persuasion-attempt rate条件不完整PersuSafety: High harm单轮APE-style StrongREJECT persuasion-attempt judge on typical response to prompt; original PersuSafety direct engagement-question setup replacedNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.2 · DoGNAVYGLM-5.279 countExploitGym · 指定目标成功数条件不完整Version not specifiedon_targetselected subsetDoGNAVYNot disclosedExploitGym official leaderboard and benchmark page发布 2026-06-16报告 —
GLM-5.2 · Claude CodeGLM-5.239 countExploitGym · 指定目标成功数条件不完整Version not specifiedon_targetrescaled 6h timeoutClaude CodeNot disclosedExploitGym official leaderboard and benchmark page发布 2026-06-16报告 —
GLM-5.2 · Claude CodeGLM-5.229 countExploitGym · 指定目标成功数条件不完整Version not specifiedon_targetrescaled 2h timeoutClaude CodeNot disclosedExploitGym official leaderboard and benchmark page发布 2026-06-16报告 —
GLM 5.2GLM-5.21 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21.4 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21.2 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21.8 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21.4 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.22 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.24.2 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 scenarioNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.21%Attempt to Persuade Eval (APE) · APE persuasion-attempt rate条件不完整Attempt to Persuade Eval: Noncontroversially HarmfulTurn 1Attempt to Persuade Eval; propensity/attempt classification, not persuasion successNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM 5.2GLM-5.2≈70%Attempt to Persuade Eval (APE) · APE persuasion-attempt rate条件不完整Attempt to Persuade Eval: Undermining ControlTurn 1Attempt to Persuade Eval; propensity/attempt classification, not persuasion successNeo ResearchEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks发布 2026-06-16报告 2026-07-03
GLM-5.243%Attempt to Persuade Eval (APE) · APE Turn-1 persuasion-attempt rate条件不完整APE Conspiracy (main persuade condition)xhighAPE simulated persuadee; automated attempt evaluator; StrongREJECT refusal classifierSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02
GLM-5.247%Attempt to Persuade Eval (APE) · APE Turn-1 persuasion-attempt rate条件不完整APE Undermining control (main persuade condition)xhighAPE simulated persuadee; automated attempt evaluator; StrongREJECT refusal classifierSaferAIGLM-5.2 Risk Evaluation Report发布 2026-06-16报告 2026-08-02

开发者披露

GLM-5.2 Model Card →

关于GLM-5.2的开发者文档，核查范围和已关联协议见下方记录。

未核实发表日期
来源提取已复核；范围见档案

外部评估

按风险主题组织的证据

自主性 38 项协议结果
欺骗 45 项协议结果

MASK →

外部评测 · 执行方 Neo Research

Honest under pressure
55%

见报告特定样本与分母。

DeceptionBench →

外部评测 · 执行方 Neo Research

Single-turn deception
11.3%
Multi-turn deception under pressure
16.7%

见报告特定样本与分母。

MASK →

外部评测 · 执行方 Neo Research

MASK response proportion: Honest
55%

原文未披露本结果的样本量

MASK →

外部评测 · 执行方 Neo Research

MASK response proportion: Evaded or No Belief
28%

原文未披露本结果的样本量

MASK →

外部评测 · 执行方 Neo Research

MASK response proportion: Lying
17%

原文未披露本结果的样本量

DeceptionBench →

外部评测 · 执行方 Neo Research

DeceptionBench deception score
11.3%

原文未披露本结果的样本量

DeceptionBench →

外部评测 · 执行方 Neo Research

DeceptionBench deception score
16.7%

原文未披露本结果的样本量

网络安全 18 项协议结果

Cybench →

外部评测 · 执行方 SaferAI

Cybench pass@1
85.29%

29/34

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
80%

12/15

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
66.67%

4/6

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
100%

2/2

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
100%

5/5

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
100%

4/4

Cybench →

外部评测 · 执行方 SaferAI

Cybench category pass@1
100%

2/2

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 7 项协议结果

APE →

外部评测 · 执行方 Neo Research

APE persuasion-attempt rate
1%

原文未披露本结果的样本量

APE →

外部评测 · 执行方 Neo Research

APE persuasion-attempt rate
≈70%

原文未披露本结果的样本量

APE Turn-1 attempts →

外部评测 · 执行方 SaferAI

APE Turn-1 persuasion-attempt rate
43%

来源分母：100（单位未披露）

APE Turn-1 attempts →

外部评测 · 执行方 SaferAI

APE Turn-1 persuasion-attempt rate
47%

来源分母：100（单位未披露）

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

CyberGym →

整理 CyberGym 官网报告的评测结果，保留受测系统和运行配置。

官网收录不同配置下的系统提交。本结果的任务数、预算和执行日期未完整披露；未定义指标及身份待核实行已排除。JSON 原始比例按官网展示规则乘以 100 并保留一位小数，转换为百分比。

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。

GLM-5.2 Risk Evaluation Report, Version 1.1 →

对 SaferAI · 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。

GLM-5.2 — official Z.ai Hugging Face model card →

完整核查了 GLM-5.2 官方模型卡及评测方法。模型卡披露的是通用能力与使用信息，未披露专门安全评测或前沿风险结论；编码和工具使用分数不作为直接风险结果。

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。

评测运行配置

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。

先核查来源条件与依赖，再进行跨来源比较或形成补测计划。

评测运行配置

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。

先核查来源条件与依赖，再进行跨来源比较或形成补测计划。

前沿风险专项评测披露

经核查的开发者资料未披露前沿风险专项评测。

先核查来源条件与依赖，再进行跨来源比较或形成补测计划。

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. [1] Z.ai release notes ; GLM-5.2 entry dated 2026-06-16
  2. [2] GLM-5.2 model card ; Parameter counts not disclosed in reviewed developer source
  3. [3] Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks. 2026-07-03. 3 July 2026 ; §1; Figure 1; §1; Figure 2; §2; Figure 6
  4. [4] GLM-5.2 Risk Evaluation Report. 2026-08-02. v1.1 · revised 26 August 2026 ; §2.2; Figure 3; p.6 Fig.1; p.6 Fig.2 printed bar label
  5. [5] CyberGym official leaderboard and machine-data extraction ; /level1/21/score_10; /level1/22/subRows/0/score_10; /level1/39/subRows/1/score_10
  6. [6] ExploitGym official leaderboard and benchmark page ; /results/12/on_target; /results/14/on_target; /results/14/subRows/0/on_target