OpenAI recently uncovered and disrupted five covert influence operations originating from Russia [5], China [1] [2] [3] [4] [5], Iran [1] [2] [3] [4] [5], and Israel in the last three months [5].


These operations involved generating content in multiple languages [3], creating fake social media profiles [3], and conducting research [3]. Threat actors attempted to use OpenAI’s language models for tasks like generating comments [5], articles [5], social media profiles [2] [3] [5], and debugging code for bots and websites [5]. The operations [1] [2] [3] [4] [5], including Spamouflage from China [2], Bad Grammar and Doppelganger from Russia [2], the International Union of Virtual Media from Iran [2], and Zero Zeno from Israel [2], were found to have low effectiveness, scoring poorly on the Brookings Institutions’ “Breakout Scale.” Despite collaborative efforts with industry partners to address AI misuse and enhance platform security, the detection of fake content online remains challenging due to the high false positive rates of existing tools [2]. OpenAI’s AI products have built-in safety defenses to prevent misuse by bad actors [1], and investigations into these operations were accelerated by AI-powered tools [1]. OpenAI thwarted campaigns related to Russia’s invasion of Ukraine [1], the conflict in Gaza [1] [3], the Indian elections [1] [4], politics in Europe and the United States [1], and criticisms of the Chinese government [1] [4]. OpenAI models were used in campaigns targeting countries like Ukraine [5], Moldova [5], the Baltics [5], the United States [1] [5], and platforms like Telegram [5], X, and [5]. The company also disrupted a commercial Israeli company using its models across social media platforms [5].


OpenAI is dedicated to detecting and mitigating such abuses at scale using generative AI [3]. Concerns about the safety of advanced AI technology persist [1], but OpenAI continues to prioritize safety and transparency in its operations [1].