AI Security AI安全 1d ago Updated 23h ago 更新于 23小时前 51

China Is Abusing AI 中国正在滥用人工智能

Chinese actors are actively exploiting existing AI tools, including American models like ChatGPT, to conduct covert propaganda and influence campaigns globally. Specific incidents include the generation of anti-AI infrastructure comics in the US and attempts to discredit foreign political figures, revealing large-scale, resource-intensive operations. AI is significantly enhancing the efficiency and scale of domestic censorship within China, allowing for faster, cheaper, and more pervasive contro 中国利用美国开源及闭源AI工具(如ChatGPT)进行隐蔽的宣传战,包括制造虚假舆论以阻碍美国AI基础设施建设。 中国执法机构被指控大规模使用AI策划针对日本政客的虚假信息活动,涉及数百名员工和数千个虚假账号。 AI技术显著提升了中国国内审查的效率与覆盖面,使信息控制更加普遍、细致且反应迅速。 训练数据中的国家主导媒体内容导致中美主流AI模型在中文语境下对中国持更正面态度,形成“观点清洗”效应。 中国本土AI模型(如DeepSeek)因法律强制要求,在政治议题上几乎总是呈现亲华立场,加剧了全球AI叙事的地缘政治风险。

75
Hot 热度
70
Quality 质量
72
Impact 影响力

Analysis 深度分析

TL;DR

  • Chinese actors are actively exploiting existing AI tools, including American models like ChatGPT, to conduct covert propaganda and influence campaigns globally.
  • Specific incidents include the generation of anti-AI infrastructure comics in the US and attempts to discredit foreign political figures, revealing large-scale, resource-intensive operations.
  • AI is significantly enhancing the efficiency and scale of domestic censorship within China, allowing for faster, cheaper, and more pervasive control over information.
  • State-sponsored Chinese news sources are passively biasing major Western AI models, causing them to produce more favorable responses about China when queried in Chinese.
  • Chinese-developed models exhibit stronger pro-state biases compared to Western counterparts due to strict regulatory mandates requiring adherence to "mainstream values."

Why It Matters

This development signals a shift in geopolitical competition where AI serves as a dual-use technology for both soft power projection and internal authoritarian consolidation. For AI practitioners and policymakers, it highlights critical vulnerabilities in model safety, training data integrity, and the potential for algorithmic bias to be weaponized by state actors. Understanding these dynamics is essential for developing robust safeguards against coordinated disinformation and ensuring the neutrality of global AI systems.

Technical Details

  • Covert Campaign Execution: Operatives utilized generative AI to create polarizing content (e.g., comics) and automate social media amplification, demonstrating the integration of LLMs into sophisticated information warfare tactics.
  • Censorship Automation: AI systems are deployed to scan vast quantities of media, automatically erasing prohibited material and flagging suspect content for human review, thereby increasing the speed and granularity of state censorship.
  • Training Data Bias: Research published in Nature indicates that heavy reliance on Chinese-language web data causes Western models like ChatGPT and Claude to generate more positive responses regarding China in Chinese compared to English.
  • Model Comparison: Comparative analysis shows Chinese models like DeepSeek align with state narratives significantly more often (99% of the time) than Western models, reflecting the impact of mandatory compliance with "positive energy" regulations.

Industry Insight

  • Data Provenance and Safety: AI developers must rigorously audit training data sources and implement stronger guardrails to detect and mitigate state-sponsored manipulation of model outputs, particularly in multilingual contexts.
  • Geopolitical Risk Management: Companies operating globally should anticipate increased scrutiny and potential bans related to AI models perceived as instruments of state propaganda, necessitating transparent alignment practices.
  • Monitoring Disinformation: The industry should invest in detection mechanisms capable of identifying AI-generated coordinated inauthentic behavior, as these tactics are becoming more sophisticated and resource-intensive.

TL;DR

  • 中国利用美国开源及闭源AI工具(如ChatGPT)进行隐蔽的宣传战,包括制造虚假舆论以阻碍美国AI基础设施建设。
  • 中国执法机构被指控大规模使用AI策划针对日本政客的虚假信息活动,涉及数百名员工和数千个虚假账号。
  • AI技术显著提升了中国国内审查的效率与覆盖面,使信息控制更加普遍、细致且反应迅速。
  • 训练数据中的国家主导媒体内容导致中美主流AI模型在中文语境下对中国持更正面态度,形成“观点清洗”效应。
  • 中国本土AI模型(如DeepSeek)因法律强制要求,在政治议题上几乎总是呈现亲华立场,加剧了全球AI叙事的地缘政治风险。

为什么值得看

这篇文章揭示了AI不仅是技术竞赛的焦点,更是地缘政治和信息战的隐形武器,提醒从业者关注模型被用于操纵舆论和审查的风险。它指出了训练数据偏见和算法合规性如何潜移默化地影响全球AI输出的中立性,对构建负责任AI体系具有警示意义。

技术解析

  • AI辅助宣传操作:攻击者利用LLM生成多语言漫画、评论及策划社交媒体运动,通过自动化手段放大特定叙事,例如针对美国数据中心建设的负面舆论。
  • 大规模虚假账号网络:情报显示存在资源密集型行动,动用数百名工作人员和数千个跨平台虚假账户,结合AI生成的伪造文件和虚假信息来压制异见。
  • 审查技术升级:AI系统被集成到内容审核流程中,能够高速扫描海量媒体数据,自动删除违规内容并标记可疑项供人工复核,大幅降低了审查成本并提高了颗粒度。
  • 数据偏见机制:研究表明,由于训练数据中包含大量新华社等官方来源,ChatGPT和Claude在处理中文关于中国政治的问题时,表现出比英文回答更积极的倾向,体现了数据源对模型输出的被动影响。
  • 合规性导致的立场偏差:中国AI模型受2022年法规约束,必须传播“正能量”并符合主流价值观,导致其在政治查询中几乎完全回避批评,呈现出高度一致的国家叙事。

行业启示

  • 警惕AI作为混合战争工具:科技公司和政策制定者需建立更严格的检测机制,识别由AI驱动的协调性虚假信息和操纵行为,防止平台被用于地缘政治渗透。
  • 重视训练数据的多样性与透明度:模型开发者应深入审计训练数据源,特别是针对特定国家或地区的语言数据,以减轻因数据不平衡导致的系统性偏见和叙事操控风险。
  • 加强全球AI治理协作:鉴于AI在审查和宣传中的双重用途,国际社会需推动关于AI伦理、透明度及跨境数据流动的标准化协议,以遏制技术被滥用为威权工具的趋势。

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。

Security 安全 Policy 政策 Ethics 伦理 Open Source 开源