Dark Box Investigation: Abu Dhabi’s AI Influence Machine — How a UAE-Directed Operation Targeted Muslim Communities in Europe and Tried to Bury Accountability for Sudan
An intelligence investigation by Anthropic has exposed a disturbing new dimension of UAE-linked influence operations: the use of artificial intelligence to build a coordinated transnational political campaign capable of generating propaganda, researching political targets, operating hundreds of inauthentic social-media identities, producing apparently independent human rights material, and attempting to influence the international debate surrounding Sudan. According to Anthropic, a single actor used its Claude AI assistant as part of an operation targeting political Islam while simultaneously seeking to influence perceptions of Abu Dhabi’s role in Sudan and the international accountability surrounding the war. Most significantly, Anthropic said it linked the operation “with high confidence” to UAE government officials, with senior Emirati officials identified as intended recipients of some of the material produced. What emerges is not the picture of an ordinary public-relations campaign, but of an organized influence architecture in which AI was used to help disguise political messaging, identify targets and manufacture the appearance of independent voices.
The scale of the operation makes the findings particularly significant. Anthropic said the actor built a network of approximately 300 inauthentic influencer accounts on social media and used Claude to research 18 members of the European Parliament as well as prominent journalists, compiling detailed personal profiles on lawmakers and reporters. Europe was therefore not simply a passive audience receiving political messaging generated elsewhere. Its political and media environment became part of the operational landscape. Individuals capable of influencing public debate were identified, researched and profiled, while a large network of apparently independent online personalities was available to amplify narratives. This combination of political targeting and manufactured social-media activity reveals how artificial intelligence can dramatically expand the capabilities of an influence operation: a relatively concentrated actor can research political figures, generate different forms of content and simultaneously create the impression that its narratives are being promoted organically by a much broader community.
Anthropic specifically described the operation’s declared mission using narrower organizational terminology, while Dark Box uses the broader analytical category of political Islam movements, parties and activists when discussing its wider political implications. The distinction matters. The available findings do not establish that Muslim communities across Europe were collectively targeted as a single group, and they should not be presented as proving that every campaign critical of political Islam is connected to Abu Dhabi. What the findings do expose, however, is a transatlantic operation targeting political Islam while profiling European lawmakers and journalists and deploying hundreds of deceptive online identities. Such an operation has obvious implications for the political and civic environment surrounding Muslim communities in Europe because narratives about political Islam, Muslim political participation, security and extremism can shape how Muslim organizations, activists and communities are treated within European political debate. The danger is therefore not merely the production of propaganda, but the possibility of foreign state-linked actors manipulating sensitive European discussions while concealing the real source of that intervention.
The Sudan component reveals an even more aggressive dimension. According to Anthropic, the operation did not simply generate material defending Abu Dhabi’s reputation. The actor compiled what the company described as “counter-accountability dossiers” on UN Special Rapporteurs who had criticized the UAE’s conduct in Sudan. This is a crucial distinction. Rather than only responding to allegations against Abu Dhabi, the operation was researching the individuals involved in scrutinizing those allegations. The people examining accountability themselves became subjects of political targeting. In the context of Sudan, where international scrutiny of foreign involvement carries potentially serious diplomatic and reputational consequences, such an operation suggests an attempt not merely to contest criticism but to map and challenge the machinery producing that criticism.
Even more disturbing was the operation’s reported use of fake civil society. Anthropic said the actor created a front NGO by copying the identity of a real Swiss organization and then published state-authored human rights reports under that borrowed identity. This strikes directly at the credibility on which international human rights advocacy depends. A government openly publishing its political position can be evaluated as government advocacy. But political material originating from a state-linked operation and presented through the identity of an apparently independent human rights organization is fundamentally different. The political origin of the message becomes concealed behind the credibility of civil society. In effect, the operation was not merely producing narratives; it was attempting to manufacture the institutional legitimacy through which those narratives would be received.
The effort extended toward the United Nations itself. Anthropic said Claude was used to ghost-write testimony intended to be delivered by two individuals before the 62nd session of the UN Human Rights Council, with specific restrictions ensuring that neither speech mentioned the UAE. That instruction is one of the most revealing details in the entire investigation because it exposes the underlying objective with unusual clarity: influence the narrative surrounding Sudan while deliberately removing Abu Dhabi from that narrative. The UAE did not simply need favorable arguments. The operation sought material capable of appearing independent while ensuring that the country whose reputation was at stake disappeared from the discussion.
This architecture depended on manufacturing independence. A fake influencer can appear to be an ordinary citizen. A front NGO can appear to represent independent civil society. State-authored material can appear to be neutral human rights documentation. Ghost-written testimony can appear to represent an autonomous voice speaking before an international institution. Hundreds of coordinated accounts can create the impression that a political narrative reflects genuine public sentiment rather than an organized campaign. The weapon is therefore not simply misinformation or propaganda. The more sophisticated weapon is false authenticity: making coordinated political influence look spontaneous, independent and credible.
Artificial intelligence provided the machinery needed to expand this model. The actor maintained an internal AI persona called “Deadshot,” supported by a master doctrine file designed to preserve the same political mission across hundreds of sessions. Claude could therefore be used to transform political doctrine and real-world developments into reports, targeting dossiers, political content and campaign material adapted to different audiences. The significance is not that AI independently invented the political agenda; the agenda came from the operator. The significance is that AI potentially made that agenda cheaper, faster and easier to reproduce across multiple formats and targets while maintaining strategic consistency.
This represents an important evolution in information warfare. Traditional influence operations require researchers to identify targets, analysts to study them, writers to produce material, political operatives to adapt narratives, and social-media networks to distribute the result. AI can compress several of those functions. It can help research targets, organize information, generate different versions of the same political argument, produce official-looking documents and maintain a common ideological framework across hundreds of interactions. When that capability is combined with approximately 300 fake influencer identities, the result is an influence system capable of making a relatively centralized operation appear decentralized and organic.
The campaigns involving political Islam and Sudan therefore reveal two sides of the same political method. On one side, adversaries can be identified, researched and surrounded by hostile narratives. On the other, Abu Dhabi can be deliberately removed from narratives that threaten its international reputation. The operator can attack the target while hiding the sponsor. It can manufacture the intermediary while concealing the political origin of the message. And it can attempt to influence not only public opinion but the politicians, journalists, civil-society organizations and international officials who help determine how that opinion is formed.
Sudan makes this model especially troubling because the influence operation approached the institutions responsible for international accountability. Compiling dossiers on UN Special Rapporteurs, copying the identity of a human rights organization and preparing testimony deliberately engineered not to mention the UAE cannot be dismissed as routine image management. The issue becomes one of protecting the integrity of the international accountability system itself. If apparently independent documentation or testimony can actually originate from a state-directed political operation, then international institutions face a much more difficult task in determining whose interests are genuinely represented by the material placed before them.
For Abu Dhabi, the potential strategic value of such an operation is obvious. The UAE does not have to win every argument openly if it can influence the environment in which the argument takes place. It does not have to place its name on every report if supposedly independent organizations can circulate favorable material. It does not have to confront every critic publicly if those critics can first be researched and profiled. And it does not have to appear at the center of every Sudan narrative if testimony can be deliberately written to exclude it.
Dark Box assesses that this is the most important lesson from Anthropic’s findings. The emerging model is not simply about using AI to produce propaganda faster. It is about using artificial intelligence to make political influence harder to identify as political influence. The real commodity being manufactured is credibility. The fake account must look authentic. The front organization must look independent. The testimony must sound autonomous. The political report must appear professionally researched. And the coordinated campaign must look like spontaneous public opinion.
There are important evidentiary boundaries. Anthropic said it could not confirm that all of the testimonies or compiled target dossiers successfully reached their intended audiences. The findings therefore establish the existence and architecture of the operation more strongly than they establish its ultimate political effectiveness. Nor do the findings justify treating Muslim communities in Europe as having been universally targeted, or every anti-political-Islam campaign as an Emirati operation. Those conclusions would exceed the available evidence. But these limitations do not diminish the central finding: Anthropic says it linked the operation with high confidence to UAE government officials, and the operation itself used AI to construct deceptive political influence infrastructure spanning social media, European political targeting, human rights narratives and the Sudan accountability debate.
For European politicians, journalists and civil-society organizations, the warning is serious. The authenticity of political campaigns can no longer be judged by appearance alone. A professional report carrying the name of an NGO may not necessarily have been produced independently. An influencer presenting themselves as an ordinary citizen may belong to a coordinated network. Apparently spontaneous waves of political commentary may have centralized financing and direction. And narratives surrounding Muslim political activism can potentially be amplified by foreign actors seeking to influence domestic European political debates.
For Sudan, the warning is even sharper. The battle over responsibility for the war is also an information war over whose evidence is believed, whose testimony reaches international institutions, which investigators retain credibility, and which foreign actors remain visible in the narrative. An operation that compiles counter-accountability dossiers while producing disguised human rights material and testimony deliberately stripped of references to the UAE is attempting to operate inside that battle rather than merely comment on it.
Dark Box concludes that Anthropic has exposed something larger than a network of approximately 300 fake accounts. It has exposed the architecture of a new kind of political influence machine. The power of that machine lies precisely in making the state behind it disappear. Abu Dhabi’s name does not need to appear on the influencer account, the human rights report, the testimony or the political narrative for those products to advance interests aligned with Abu Dhabi.
That is what makes the operation so dangerous. Propaganda becomes more powerful when it no longer looks like propaganda. State influence becomes harder to challenge when it looks like independent civil society. Political targeting becomes more difficult to identify when it is disguised as ordinary research. And an international influence operation becomes more effective when hundreds of artificial voices create the illusion that no operation exists at all.
Anthropic disrupted the network it discovered. But the exposure of “Deadshot” leaves a much more uncomfortable question behind: if a UAE-directed actor was already using artificial intelligence to target political adversaries, profile European politicians and journalists, manufacture supposedly independent voices and influence the accountability battle surrounding Sudan, how many similar operations remain hidden?
The battlefield has changed. Artificial intelligence is no longer merely a communications tool. It is becoming infrastructure for political influence, narrative laundering and synthetic authenticity. And according to Anthropic’s investigation, a UAE-directed actor was already attempting to use that infrastructure to target political Islam, penetrate the European information environment, and push Abu Dhabi out of the line of fire over Sudan.
