与中国有关的影响力行动正瞄准美国AI辩论

我们的使命是确保通用人工智能惠及全人类。我们通过部署创新来推进这一使命，构建民主化的人工智能：即由民主原则塑造、受常识规则约束、旨在帮助人们解决难题同时保护其免受实际伤害的人工智能。这一使命还要求识别并阻止威权政权及其代理人利用人工智能系统胁迫批评者、监控社群或秘密干预民主社会的企图。

在本报告中，我们描述了两个可能源自中国的ChatGPT账户集群。这些账户在支持明显隐蔽影响力行动时使用我们的模型，试图操纵关于美国人工智能及更广泛科技政策的合法辩论，我们已将其封禁。

第一个集群生成社交媒体评论和图片，声称人工智能数据中心建设导致普通家庭电价上涨。我们将该集群命名为"数据中心跟风"行动。

第二个集群生成评论和图片，批评美国关税试图主导技术竞争，并在提示词中明确要求输出内容不得包含中国领导人习近平，仅可提及特朗普总统。该集群与一个可能由虚假社交媒体账户组成的网络存在关联，该网络还曾通过声称ChatGPT用户数据遭泄露来针对OpenAI。这些指控完全虚假。我们将第二个集群命名为"技术与关税"行动。

针对OpenAI和美国数据中心建设的攻击之所以值得关注，并非因为该行动似乎改变了公众舆论，而是因为它表明源自中国的影响力操作者正在测试针对人工智能基础设施的叙事——而这是美国技术领导地位、经济增长及更广泛民主人工智能生态系统的基石。该行动试图利用并放大公众对能源价格及数据中心开发本地影响的现有担忧，但我们未发现其活动范围超出自身范畴的重大突破证据。外国影响力行动长期试图依附现有本地议题和真诚信念，借此建立可信度、加剧分歧或激化公众不信任。在此案例中，操作者试图秘密介入美国关于国家人工智能能力未来的合法辩论，同时隐藏自身身份与动机。

通过发布这些发现，我们旨在帮助行业、政府、公民社会及公众更好地识别并阻止外国威胁行为者操纵合法公共辩论、削弱民主制度、推进具有人工智能特征的极权主义——即利用人工智能进行监控、审查及对政治、社会和私人生活的控制。

Our mission is to ensure that artificial general intelligence benefits all of humanity. We advance this mission by deploying our innovations to build democratic AI: AI shaped by democratic principles, governed by common-sense rules and designed to help people solve hard problems while protecting them from real harm. That mission also requires identifying and disrupting attempts by authoritarian regimes and their proxies to use AI systems to coerce critics, surveil communities or covertly interfere in democratic societies.

In this report, we describe two clusters of ChatGPT accounts likely originating from China that we banned after they used our models in support of apparent covert influence operations that promoted narratives in an attempt to manipulate a legitimate debate about American AI and wider tech policies.

The first cluster generated social media comments and images claiming that data center buildouts for AI were increasing electricity prices for average families. We named this cluster the “Data Center Bandwagon” campaign.

The second cluster generated comments and images criticizing US tariffs as attempts to dominate technological competition and specified in their prompts that the content should not include China’s leader Xi Jinping in the output and instead include only President Trump. This cluster was connected to a network of likely inauthentic social media accounts that were also likely targeting OpenAI by claiming ChatGPT user data had been compromised. These allegations were entirely false. We named this second cluster the “Tech and Tariffs” campaign.

The targeting of OpenAI and US data center buildouts is significant not because the operation appears to have shifted public opinion, but because it shows PRC-origin influence operators testing narratives against AI infrastructure – a foundation of US technological leadership, economic growth and the broader democratic AI ecosystem. The operation sought to exploit and amplify existing public concerns about energy prices and local impacts of data center development, but we found no evidence of meaningful breakout beyond its own activity. Foreign influence operations have long sought to latch onto existing local issues and sincerely held beliefs, using them to build credibility, amplify divisions or exacerbate public distrust. In this case, the operators attempted to covertly insert themselves into an ongoing American debate about the future of the country’s AI capabilities while hiding who they were and what motivated them.

By publishing these findings, we aim to help our industry, governments, civil society and the public better identify and disrupt attempts by foreign threat actors to manipulate legitimate public debates, weaken democratic institutions and advance totalitarianism with AI characteristics - the use of AI for surveillance, censorship and control over political, social and private life.