AI News AI资讯 3h ago Updated 47m ago 更新于 47分钟前 48

Fake US thinktank set up and funded by Israel sought to game AI for propaganda 以色列资助伪造美国智库试图操纵AI传播宣传

A fake US thinktank called the Hanover Institute for Public Policy was created and funded by Israel to produce pro-Israel content optimized for AI chatbot citation The site published 124 reports totaling over 560,000 words in just nine days (August 6-14), all framed as neutral academic research The content was built on Res, a commercial "AI-native content platform" specializing in generative engine optimization to get cited by ChatGPT, Perplexity, Claude, and Gemini Piro Inc, a New York media co 以色列通过虚假智库"Hanover Institute for Public Policy"在9天内发布124份亲以报告(超56万字),旨在操纵ChatGPT等AI聊天机器人传播其宣传叙事 项目由纽约公司Piro Inc运营,使用Res平台的"生成式引擎优化"技术,专门优化内容以被AI系统引用,已按FARA向美国司法部备案 报告采用伪学术形式,大量引用世界银行、联合国、大赦国际等权威来源,但通过选择性框架(如反复强调"法院尚未裁决")淡化以色列战争罪指控 同一项2022年Facebook反犹主义研究被引用454次横跨88份报告,显示系统性的内容操纵策略 该行动反映以色列在美国公众舆论下滑背景下

72
Hot 热度
68
Quality 质量
62
Impact 影响力

Analysis 深度分析

TL;DR

  • A fake US thinktank called the Hanover Institute for Public Policy was created and funded by Israel to produce pro-Israel content optimized for AI chatbot citation
  • The site published 124 reports totaling over 560,000 words in just nine days (August 6-14), all framed as neutral academic research
  • The content was built on Res, a commercial "AI-native content platform" specializing in generative engine optimization to get cited by ChatGPT, Perplexity, Claude, and Gemini
  • Piro Inc, a New York media company, filed the site under the Foreign Agents Registration Act (FARA) as material distributed for the Israeli government
  • The reports used quasi-academic formatting with citations from credible sources but employed framing tactics to blunt criticism of Israel, including repeated claims that "no court has entered a verdict" on genocide/apartheid allegations

Why It Matters

This represents a significant case study in how foreign governments can exploit AI optimization techniques to manipulate information ecosystems and shape chatbot outputs at scale. It raises urgent questions about content provenance, AI training data integrity, and the vulnerability of generative systems to coordinated propaganda campaigns.

Technical Details

  • The Hanover Institute site initially carried a standard llms.txt file from Res (an AI-native content platform), which was later replaced with a bespoke version after Guardian exposure on August 13
  • Res founder Hai Tran registered as a subcontractor to Piro Inc under FARA on July 22, listing his role as "digital strategist"
  • Piro Inc advertises a service called "AI Story Optimization" that authors content "engineered for how LLMs evaluate credibility"
  • The 124 reports averaged approximately 4,500 words each, with a burst of 73 reports (354,000 words) published on August 12-13 alone
  • Headlines were deliberately structured as chatbot-style questions (e.g., "Is Israel Carrying Out a Genocide in Gaza?") to match common user queries
  • Citation patterns revealed manipulation: a single 2022 study on antisemitic Facebook comments was cited 454 times across 88 reports on unrelated topics

Industry Insight

  • AI companies must develop more robust content provenance and source verification systems to detect and resist coordinated optimization campaigns designed to game their training and retrieval pipelines
  • The rise of "generative engine optimization" as a commercial service creates new vectors for state-sponsored influence operations that exploit the trust AI systems place in structured, well-cited content
  • Researchers and practitioners should treat AI-optimized content platforms with heightened skepticism, particularly when sources lack verifiable institutional backing, physical addresses, or named staff

TL;DR

  • 以色列通过虚假智库"Hanover Institute for Public Policy"在9天内发布124份亲以报告(超56万字),旨在操纵ChatGPT等AI聊天机器人传播其宣传叙事
  • 项目由纽约公司Piro Inc运营,使用Res平台的"生成式引擎优化"技术,专门优化内容以被AI系统引用,已按FARA向美国司法部备案
  • 报告采用伪学术形式,大量引用世界银行、联合国、大赦国际等权威来源,但通过选择性框架(如反复强调"法院尚未裁决")淡化以色列战争罪指控
  • 同一项2022年Facebook反犹主义研究被引用454次横跨88份报告,显示系统性的内容操纵策略
  • 该行动反映以色列在美国公众舆论下滑背景下,转向利用AI技术进行新型公共外交和宣传战

为什么值得看

这篇文章揭示了AI时代信息操纵的新形态——通过"生成式引擎优化"技术系统性影响AI聊天机器人的回答,为AI从业者提供了关于AI内容生态、信息可信度和技术伦理的重要警示。

技术解析

  • 生成式引擎优化(GEO)技术:使用llms.txt文件向AI系统声明内容结构和可读性,优化网站被ChatGPT、Perplexity、Claude、Gemini等模型引用的概率,这是传统SEO在AI检索时代的演进形态
  • 规模化内容生产:9天内发布124份报告,平均4500字/篇,高峰期2天产出73份近35万字,采用标准化模板快速批量生产,标题设计模仿用户向聊天机器人提问的句式
  • 引用操纵策略:大量引用权威机构数据但采用选择性框架,如20份报告反复使用"无法院裁决"表述淡化战争罪指控,55份报告使用"事件参与方" caveat模糊伤亡数据
  • 重复引用模式:同一项2022年英法德媒体Facebook反犹主义研究被引用454次横跨88份报告,即使在与土地登记、加沙死亡人数等无关主题中也被强行植入
  • 技术痕迹管理:最初使用Res平台标准化llms.txt文件,被发现后替换为定制版本并移除平台标识,显示对技术暴露的敏感性

行业启示

  • AI内容生态脆弱性:大型语言模型依赖公开网络内容进行检索和引用,恶意行为者可通过批量生产优化内容系统性污染AI输出,AI平台和研究者需开发内容溯源和可信度验证机制
  • GEO成为新型宣传战场:生成式引擎优化正在取代传统SEO成为信息战新工具,AI从业者需关注"AI原生宣传内容"的检测方法,平台需建立针对GEO操纵的防御机制
  • 监管框架滞后:该项目通过FARA备案披露资金源头,但实际运作高度隐蔽,反映出现有监管在应对AI时代跨国信息操纵时的不足,需要技术平台、监管机构和学术界协同建立新的透明度标准

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Security 安全 Alignment 对齐 Ethics 伦理 Policy 政策