AINEWS
The AI lab's threat report describes impersonation of two Sudanese rights workers to plant UAE aligned material at the Human Rights Council.
Anthropic says a UAE-linked operation used Claude, its commercially available AI assistant, to impersonate two named individuals from a real Sudanese human rights organization and draft testimonies for the 62nd session of the UN Human Rights Council under instructions that banned any mention of the UAE. The AI lab, which said it identified the operation "with high confidence," published the findings in a threat report on September 11, 2026. The UAE did not respond to AFP requests for comment.
The operation targeted the evidentiary pipeline of the world's main human rights forum during an active mass-atrocity conflict. Aid workers estimate that more than 200,000 people have been killed in Sudan and that one million have been displaced. The 62nd session of the Human Rights Council, the principal UN body that reviews country human rights records, is one of the few international venues where the war is documented in real time. Testimony submitted under the council's procedures is treated as the work of independent civil society; state messaging is not. Anthropic's report says the operation was built to break that distinction.
According to Anthropic, the operation used Claude to generate reports, written materials, witness testimonies, and social media content about Sudan. Operatives impersonated two people affiliated with a genuine Sudanese rights organization, so that UAE-aligned material could reach the UN as if it came from independent local witnesses rather than from a party to the conflict. Drafted testimonies for the council's 62nd session were written under "specific restrictions" intended to scrub any reference to the UAE. The materials generated by the operation, in Anthropic's account, were not generic propaganda. They were shaped to look like first-person witness statements from a recognized Sudanese civil society source.
The persona at the center of the campaign was called "Deadshot" in Anthropic's report. According to the lab, Deadshot was directed to run what the company described as a "coordinated transatlantic and regional operation to dismantle the Muslim Brotherhood globally." The Sudan-focused track sat inside that broader mandate, and the impersonation of named Sudanese witnesses was its instrument. Anthropic's primary target identification was the Muslim Brotherhood; the UN testimony track was the secondary line of attack, aimed at experts who have criticized the UAE's role in the Sudan war.
The disclosure matters because it is the first detailed public attribution of a commercial AI tool to a campaign aimed at the evidentiary record of a multilateral human rights body during a live war. Wire coverage of the Anthropic report has been carried by France 24, Al-Monitor, Middle East Eye, NDTV, RTL, and regional outlets including Albawaba. The September 11 release put the report at the center of a coordinated news cycle. Al-Monitor said it also reached out to the UAE Foreign Ministry for comment; the UAE has not publicly addressed the report.
Anthropic's confidence language is its own. The lab's "with high confidence" UAE attribution is a vendor assessment, not an independent intelligence finding, and independent corroboration of the impersonation is currently limited to outlet repetition of the report. Middle East Eye notes that the UAE "continues to deny its involvement in the war, despite mounting evidence reported on by Middle East Eye, human rights organisations and the UN itself." No UAE official quoted in the September 11 coverage has disputed the specifics of the Anthropic report; the silence is the response on the record.
The next test is the council's own record. The 62nd session convenes in September 2026 in Geneva. If the fabricated testimonies were filed, they should appear in the council's submissions log, and the two impersonated Sudanese workers will be able to confirm or deny that material was filed in their names. Anthropic's report gives reporters and the rights organization concerned a verifiable artifact: the names of the impersonated individuals, the AI persona that produced the text, and the session that the text was aimed at. That is enough to confirm or refute the operation without relying on the lab's confidence language alone.