风险治理 · 研究方法

Petri : An open-source auditing tool to accelerate AI safety research

NUWA 研究解读 · 独立复核待完成

首次发表: 2025-10-06来源核验: 2026-10-25

原始来源 ↗引用本记录 ↓

NUWA Research Note

研究问题研究设计与假设来源发现 NUWA 解读局限与相反证据研究联系研究与决策意义机构与角色原始资源

模型辅助整理 · 选读原始全文或官方文章中的相关章节；尚未完成逐章节审计。

研究速读

自动多轮审计可产生假设与轨迹，但试点分数仍属暂定。

研究问题

种子指令驱动的模拟互动能否高效发现值得深入调查的对齐行为？

研究设计与假设

审计智能体通过模拟用户与工具规划和交互，裁判对轨迹评分；试点覆盖多种种子，并对举报行为做消融。

来源发现

举报随授权自主性、领导共谋及所感知问题性质改变。

一些无害问题叙事也引发披露，使单纯最小化伤害解释受到质疑。

NUWA 解读

NUWA 解读：阳性轨迹与受控消融比把聚合试点分数视为通用对齐质量更有信息。

局限与相反证据

文章明确说明场景集合有限、AI 审计者不完美且存在评估意识问题；本次未阅读单独技术报告。

研究联系

NUWA 关联：自动对齐缓解在基准优化后使用 Petri 补充审计，但并不构成穷尽安全测试。

研究与决策意义

NUWA 建议：检查轨迹并改变假定原因，分数旁保留种子覆盖与裁判局限。

精确来源位置

Petri : An open-source auditing tool to accelerate AI safety research · Broad-coverage pilot-alignment evaluationsSelected original resource https://www.anthropic.com/research/petri-open-source-auditing; sha256:7e988b2e577de252b4b8404d10266c39adfc5af9a52b8c8555d6022c1e7d2650; archived source capture; exact capture time not independently established
Petri : An open-source auditing tool to accelerate AI safety research · Case study: Whistleblowing behaviorSelected original resource https://www.anthropic.com/research/petri-open-source-auditing; sha256:7e988b2e577de252b4b8404d10266c39adfc5af9a52b8c8555d6022c1e7d2650; archived source capture; exact capture time not independently established
Petri : An open-source auditing tool to accelerate AI safety research · Get startedSelected original resource https://www.anthropic.com/research/petri-open-source-auditing; sha256:7e988b2e577de252b4b8404d10266c39adfc5af9a52b8c8555d6022c1e7d2650; archived source capture; exact capture time not independently established

来源核查、数据提取、编辑批准与实验复现分别记录。方法说明 →

机构与角色

发布机构 Anthropic

评测协议

Petri →

引用与复用

引用研究发现时请引用原始出版物；引用 NUWA 的整理与解释时，单独引用本站记录及对应快照。

BibTeX ↓RIS ↓记录与来源元数据 · JSON ↓阅读覆盖 · JSON 提交更正

原始出版物 · BibTeX

@misc{petri,
  title = {Petri : An open-source auditing tool to accelerate AI safety research},
  url = {https://www.anthropic.com/research/petri-open-source-auditing},
  year = {2025},
  date = {2025-10-06},
  version = {Source capture 2026-10-25},
  note = {Source capture 2026-10-25}
}

NUWA 记录 · BibTeX

@misc{petri-nuwa,
  title = {Petri : An open-source auditing tool to accelerate AI safety research — NUWA research record},
  url = {https://whitzard.tech/nuwa/whitzard-index/work/petri/},
  version = {2026-10-29.1},
  organization = {NUWA Frontier AI Safety Lab},
  howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--petri.json},
  note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--petri.json.}
}

证据快照 2026-10-29.1 · 索引更新 · RSS

参考文献

[1] Petri : An open-source auditing tool to accelerate AI safety research. 2025-10-06 ; Official publication masthead and article byline; Broad-coverage pilot-alignment evaluations; Case study: Whistleblowing behavior