Anthropic has made public in a report that it has managed to neutralize a disinformation campaign aimed at discrediting the Muslim Brotherhood and shaping international perception about the role of the United Arab Emirates in the war in Sudan, noting that the operation would have been orchestrated by "high Emirati officials."
According to the document, a single operator, identified as 'Deadshot' and working from his own private platform, employed his AI assistant to repeat the same sequence of instructions in numerous parallel sessions, with the purpose of sustaining a coordinated operation on a regional and transatlantic scale.
The report details that "They built and managed a network of approximately 300 invented 'influencers' on social media accounts. They created a front NGO that copied the identity of a real Swiss organization and published state-authored Human Rights reports in its name."
Furthermore, it indicates that "official testimonies were drafted to be delivered to two people at the 62nd session of the United Nations Human Rights Council." "The content was designed with specific restrictions to ensure that neither of the two speeches mentioned the UAE," the text specifies.
The company adds that a thorough investigation was carried out "on 18 members of the European Parliament and prominent journalists," for whom exhaustive dossiers were prepared, and that files were also developed on UN special rapporteurs who had criticized Abu Dhabi's actions in Sudan.
The report emphasizes that "Although the operation was constructed in such a way that it could not be traced back to actors, our investigation has securely linked it to officials of the UAE government."
ABOUT THE CONFLICT IN YEMEN
In the same publication, Anthropic explains that it detected "a cell of actors" located in northern Yemen —without directly mentioning the Houthi rebels— that was carrying out three weapons development programs with the help of its assistant, including a guided rocket, a ballistic missile with a range exceeding 2,000 kilometers, and a missile from the R2000 series.
The report states that "The actors used Claude Code instead of human software engineers to develop the guidance, navigation, and control (GNC) software that directs and stabilizes a flying vehicle. For example, they used Claude to integrate an open-source autopilot into a flight computer the size of a phone, writing the control and position estimation software."
The company clarifies that, in most cases, the protection systems built into its model "blocked many of their requests," although not "all" of them. "They employed various tactics to circumvent these measures, such as hiding their targets and the products for which the software was intended, and fragmented their work into multiple sessions so that none of them revealed their true intentions on their own," it adds.
Finally, Anthropic points out that the accounts involved were suspended and that they shared the findings "with partners from the public and private sectors to mitigate risks." "We have evidence that the actors had already created a set of simulation tools outside that do not depend on Claude," concludes the company.