Anthropic said two covert campaigns used its Claude AI system to shape political debate, linking one with high confidence to United Arab Emirates government officials and the other to an Iranian opposition network.
Its September threat intelligence report covers activity disrupted between December 2025 and August 2026. The UAE-linked campaign maintained roughly 300 inauthentic influencer social media accounts; the other used AI to pose as a real-world activist when contacting people he knew.
Anthropic described the UAE effort as the work of one operator who used Claude to run Deadshot, an AI persona on a platform the operator controlled. The campaign also copied a Swiss organization’s identity for a front group and researched 18 members of the European Parliament and prominent journalists.
The operator gathered material on United Nations special rapporteurs who had criticized the UAE’s conduct in Sudan, Anthropic said. The operator also drafted statements for two people to present at the United Nations Human Rights Council’s 62nd regular session, held June 15 to July 8, 2026. Anthropic said it could not confirm whether the statements or the material on the rapporteurs reached their intended audiences, or whether the campaign affected policy.
A U.N. fact-finding mission reported in September 2025 that both the Sudanese Armed Forces and the Rapid Support Forces had carried out large-scale attacks on civilians.
In the other case, Anthropic linked the activity to the People’s Mojahedin Organization of Iran (PMOI/MEK) and its political front, the National Council of Resistance of Iran (NCRI). The company said the campaign sought to strengthen the MEK’s position within Iran’s opposition.
Operators had a shared AI agent named Viktor review roughly 8,400 Telegram posts by a real-world activist, then used it to hold political conversations with his contacts. To Anthropic’s knowledge, those contacts did not realize AI was assisting the account.
The network examined more than 500 social media channels and approximately 51,944 archived messages to assemble profiles of dozens of people in Iran. Those profiles included political views and arrest histories. Anthropic said at least four individuals running the campaign worked for official NCRI media outlets.
Anthropic said it could not determine how tightly the MEK/NCRI campaign was directed or how much genuine engagement its accounts drew. It said it banned accounts connected with both campaigns.
