Leading artificial intelligence firm Anthropic has announced the discovery and disruption of a coordinated, UAE-directed influence operation that exploited its Claude AI assistant to shape global discourse on two high-stakes regional issues: the status of the Muslim Brotherhood and the ongoing conflict in Sudan. The details of this covert campaign are laid out in the company’s September 2024 intelligence threat report, marking one of the first documented cases of a state-aligned actor abusing commercial generative AI to advance covert geopolitical goals.
According to Anthropic’s investigation, the operation was run by a single actor working on behalf of United Arab Emirates government officials, with high confidence linking the work to senior Emirati decision-makers. The core mission laid out in the operation’s internal guiding documents, which the actor fed directly into Claude across hundreds of work sessions, was to execute a coordinated transatlantic and regional campaign aimed at dismantling the global Muslim Brotherhood movement. Beyond that, the operation also sought to manipulate global public opinion regarding the UAE’s well-documented involvement in Sudan’s two-year civil war.
The UAE has publicly long denied any role in the Sudan conflict, but multiple independent investigations from Middle East Eye, international human rights groups, and United Nations researchers have confirmed that Abu Dhabi provides extensive military and political backing to the Rapid Support Forces (RSF), the Sudanese paramilitary group that has been widely accused of perpetrating genocide and mass atrocities in the Darfur region. The influence operation’s secondary goal was specifically designed to undermine international efforts to hold the UAE accountable for its support for the RSF at the UN.
To carry out this campaign, the actor leveraged Claude’s generative capabilities to build an extensive infrastructure of deceptive content and networks. At its core, the operation built a network of roughly 300 inauthentic influencer accounts across major social media platforms, all designed to amplify the operation’s narratives to broader audiences. The actor also used Claude to create a fake non-governmental organization that stole the identity of a legitimate, existing Swiss human rights group, publishing state-aligned “human rights reports” under the stolen organization’s legitimacy.
One of the operation’s most high-stakes plans targeted the UN Human Rights Council’s 62nd session. Claude was used to ghost-write official testimonies that were intended to be delivered by two speakers at the session, with explicit instructions built into the content that the UAE could not be mentioned in any critical context. Beyond institutional targeting, the actor used Claude to conduct open-source research and build detailed personal profiles of 18 members of the European Parliament and prominent international journalists who cover Middle East affairs. The operation also compiled dedicated counter-accountability dossiers targeting UN Special Rapporteurs who had previously published criticism of the UAE’s conduct in Sudan. All of the generated content, from cloned identity documents to profiled target lists, was prepared for direct delivery to senior UAE government officials, Anthropic confirmed.
The actor operated a custom private AI persona codenamed “Deadshot” on their own private infrastructure to coordinate the campaign, embedding a master mission file that instructed Claude to reiterate the operation’s core goals across all work sessions. The standardized workflow mixed real-world event data and the actor’s pre-written political doctrine, then used Claude to refine the content into official-looking intelligence reports, testimonies, and dossiers. The UAE-linked actor also directly funded the social media network created to amplify the operation’s content to wider global audiences.
To date, Anthropic notes that it has not been able to confirm whether any of the prepared testimonies or compiled targeting dossiers successfully reached their intended audiences at the UN or within the UAE government. The UAE has not yet issued a public response to Anthropic’s findings, and continues to maintain its official position that it has no involvement in the Sudan conflict. The disclosure shines a new light on the growing risk of state actors misusing commercial generative AI tools to conduct covert geopolitical influence operations, undermining multilateral accountability efforts and manipulating global public discourse.