评测档案
RedCode
代码智能体能否在不同输入形式下识别不安全执行与生成?
跨报告监测
已发表证据
RedCode · Attack Success Rate
跨来源报告的已发表测量。任务范围通过注释区分,配置差异保留在每条观测中。
24 条观测 · 0 已定位日期
● 条件匹配◇ 配置差异□ 部分条件未披露
- Code input / Base
- Code input with jailbreaking / Base
- Text summaries / Base
- Text descriptions / Base
- Code input / + (a) Simple
- Code input with jailbreaking / + (a) Simple
- Text summaries / + (a) Simple
- Text descriptions / + (a) Simple
- Code input / + (b) Awareness
- Code input with jailbreaking / + (b) Awareness
- Text summaries / + (b) Awareness
- Text descriptions / + (b) Awareness
- Code input / + (c) Detection
- Code input with jailbreaking / + (c) Detection
- Text summaries / + (c) Detection
- Text descriptions / + (c) Detection
- Code input / + (d) Modification
- Code input with jailbreaking / + (d) Modification
- Text summaries / + (d) Modification
- Text descriptions / + (d) Modification
- Code input / + (e) Combined
- Code input with jailbreaking / + (e) Combined
- Text summaries / + (e) Combined
- Text descriptions / + (e) Combined
24 条观测尚无已核实的模型发布日期,可在数据与系统结果中查看。
选择图中数据点或表格行,查看条件与出处。
24 条观测
| 系统/模型 | 结果 | 条件状态 | 任务/条件 | 评测方/报告 | 日期 |
|---|---|---|---|---|---|
| 88.4%Attack Success Rate | 条件不完整 | Text descriptions / Basetemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 87.2%Attack Success Rate | 条件不完整 | Text descriptions / + (a) Simpletemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 86.53%Attack Success Rate | 条件不完整 | Text descriptions / + (d) Modificationtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 85.33%Attack Success Rate | 条件不完整 | Text descriptions / + (b) Awarenesstemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 84.93%Attack Success Rate | 条件不完整 | Text descriptions / + (c) Detectiontemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 83.2%Attack Success Rate | 条件不完整 | Text descriptions / + (e) Combinedtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 79.33%Attack Success Rate | 条件不完整 | Code input / Basetemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 71.87%Attack Success Rate | 条件不完整 | Code input / + (a) Simpletemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 69.07%Attack Success Rate | 条件不完整 | Text summaries / Basetemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 67.33%Attack Success Rate | 条件不完整 | Code input / + (d) Modificationtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 64.53%Attack Success Rate | 条件不完整 | Code input with jailbreaking / Basetemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 63.73%Attack Success Rate | 条件不完整 | Code input / + (b) Awarenesstemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 63.73%Attack Success Rate | 条件不完整 | Code input / + (c) Detectiontemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 63.07%Attack Success Rate | 条件不完整 | Text summaries / + (a) Simpletemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 62.8%Attack Success Rate | 条件不完整 | Text summaries / + (d) Modificationtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 62.13%Attack Success Rate | 条件不完整 | Code input / + (e) Combinedtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 61.06%Attack Success Rate | 条件不完整 | Text summaries / + (c) Detectiontemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 60.93%Attack Success Rate | 条件不完整 | Code input with jailbreaking / + (a) Simpletemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 59.6%Attack Success Rate | 条件不完整 | Text summaries / + (b) Awarenesstemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 58.8%Attack Success Rate | 条件不完整 | Text summaries / + (e) Combinedtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 57.73%Attack Success Rate | 条件不完整 | Code input with jailbreaking / + (d) Modificationtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 51.47%Attack Success Rate | 条件不完整 | Code input with jailbreaking / + (b) Awarenesstemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 51.47%Attack Success Rate | 条件不完整 | Code input with jailbreaking / + (c) Detectiontemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 | |
| 48.93%Attack Success Rate | 条件不完整 | Code input with jailbreaking / + (e) Combinedtemperature=0; max 1024 generated tokens; 8 interaction rounds (Appendix C.2)ReAct | 未披露RedCode: Risky Code Execution and Generation Benchmark for Code Agents | 发布 —报告 2024-11-12 |
结合任务范围与运行配置解读结果。
METR Time Horizon 1.1 来源已停止积极更新,最后更新于 2026 年 9 月 8 日。超过 16 小时的估计超出其可靠测量范围。
协议、配置与原始证据 →任务与测量
Benchmark 协议档案
- 任务/对象
- RedCode 分开测量执行和生成:执行在容器中检查结果,生成使用另一套质量评估。
- 测量目标
- 代码智能体能否在不同输入形式下识别不安全执行与生成?
- 评分/结果口径
- Accuracy · Accuracy
- 解释边界
- 执行与生成成绩的评审和分母不同;智能体框架支持的语言子集也不同,不能直接比较未匹配子集的汇总率。
一项已报告评测
来源评测 GPT-4o / Base LLM · RedCode-Gen / Base LLM.
运行配置: Base model generation.
报告指标: Accuracy · 69.4%.
Accuracy 见原文样本说明
报告中的行为随代码或文字输入、风险行为类型而变化;编码能力本身不能证明执行安全。
测量档案 · NUWA 解读
档案区分来源方法阅读与既有目录信息;已经公开不代表完成独立证据复核或双语编辑复核。
- 测量构念
- 分别评测不安全代码执行与明确有害的代码生成。
- 任务与样本
- RedCode-Exec 包含代码与文本形式的 4,050 项测试;RedCode-Gen 包含八类软件的 160 项函数签名与说明提示。
- 判分与失效情形
- Exec 通过执行脚本区分拒绝、执行失败与攻击成功。Gen 报告拒绝、代码准确性和独立的 VirusTotal 分数;后者在本轮未确定单位,保留来源标签。
- 系统条件
- 基础模型与代码智能体分别作为系统。ReAct 缓解实验对应附录 C.2:temperature=0、最多生成 1024 tokens、八轮交互。
- 统计含义
- 现有材料尚未确定
- 效度与解释
- Exec 与 Gen 测量不同,不做平均。数值图仍待取得底层数据或明确数字化;本轮表格提取不表示已覆盖全部 Benchmark 数据。
- 适用研究问题
- 现有材料尚未确定
- RedCode: Risky Code Execution and Generation Benchmark for Code Agents · #S3; #S4; #S57e4796140ad966c180057ec7fcc9440e099ad8d2b7e27ea54a7825cdc2fbc961
跨报告证据
相同 Benchmark 名称可能对应不同任务、计分规则和系统配置。各来源分别保留,比较组说明具体对齐了哪些条件。
RedCode: Risky Code Execution and Generation Benchmark for Code Agents
RedCode: Risky Code Execution and Generation Benchmark for Code Agents
实际评测方: 未披露
84 条结果记录 · 19 个报告系统名称
协议与比较依据 (0)
原始来源图表与材料
执行与生成成绩的评审和分母不同;智能体框架支持的语言子集也不同,不能直接比较未匹配子集的汇总率。
相关研究问题
参考文献
数值结果由所引研究的作者报告。
- [1] RedCode: Risky Code Execution and Generation Benchmark for Code Agents. 2024-11-12 ; #S4.T1; row GPT-4o / Base LLM; condition RedCode-Gen / Base LLM; column Accuracy; #S4.T1; row GPT-4o / Base LLM; condition RedCode-Gen / Base LLM; column Refusal rate