Anthropic Bans Accounts Linked to Surveillance and Influence Operations in Middle East
Anthropic has taken significant action against accounts linked to surveillance and influence operations in the Middle East. In June 2026, the company banned an account used to create a surveillance platform that classified social media users in Iran and the Persian Gulf based on political loyalties and demographics. This account was associated with S2T Unlocking Cyberspace, a firm previously flagged for similar activities.
According to Anthropic's threat intelligence report published on September 10, 2026, the banned account had amassed over 255 counterfeit social media personas in multiple languages, designed to infiltrate both pro- and anti-government communities in Iran. Although Anthropic's safeguards blocked explicit profiling requests, some surveillance-related tools managed to bypass these protections.
In a separate incident, the report detailed a UAE-directed influence operation, tracked as GTG-84002, which utilized Claude to target the Muslim Brotherhood globally. Managed by an individual identified as “Deadshot,” this campaign involved approximately 300 inauthentic social media accounts, a fabricated NGO, and ghostwritten testimony submitted to the UN Human Rights Council. Anthropic has high confidence that this operation was linked to UAE government officials.
Deadshot's activities included developing personal dossiers on European Parliament members and journalists, aimed at identifying pressure points within European political institutions. The operation's objective aligns with the UAE's long-standing foreign policy against the Muslim Brotherhood, which it designated as a terrorist organization in 2014.
Anthropic's report highlights a broader trend of misuse of its AI tools, with similar operations previously identified by OpenAI. In response to these findings, Anthropic has banned all accounts associated with these campaigns and implemented additional safeguards to prevent future misuse.
Updated 15:01 UTC
New Developments on Anthropic's AI Misuse
On September 10, 2026, Anthropic reported that a weapons-engineering cell in northern Yemen had been using its Claude AI models for missile guidance and control software development for approximately nine months. This group, likely connected to Iran-backed Houthi militants, operated from December 2025 to August 2026 before their accounts were banned.
The cell was engaged in three distinct weapons programs, including a guided rocket utilizing a phone-grade flight computer, a multi-stage ballistic missile with ranges over 2,000 km, and plans for a hypersonic glide vehicle within the R2000 missile family.
Anthropic's report detailed that the Yemeni operatives ran multiple instances of Claude AI in parallel for coding, research, and output review. Despite the detection and banning of accounts, the group had developed an offline simulation toolkit to continue their work.
This incident is part of a broader trend, with Anthropic documenting six separate misuse incidents involving weapons software development linked to state actors in China and Russia. Notably, no operational weapons were confirmed to have emerged from the Yemeni cell's projects.
Anthropic's decision to publicly share these findings highlights the challenges AI companies face in ensuring safety and security against determined adversaries.
FAQ
What actions did Anthropic take against accounts linked to surveillance operations?
Anthropic banned accounts associated with surveillance and influence operations in the Middle East, including one that created a surveillance platform for classifying social media users in Iran and the Persian Gulf.
What was the purpose of the banned account related to S2T Unlocking Cyberspace?
The banned account was used to create a surveillance platform that classified social media users based on political loyalties and demographics, targeting both pro- and anti-government communities in Iran.
What was the GTG-84002 operation mentioned in the report?
The GTG-84002 operation was a UAE-directed influence campaign that used AI tools to target the Muslim Brotherhood globally, involving approximately 300 inauthentic social media accounts and a fabricated NGO.
Who was responsible for managing the GTG-84002 operation?
The operation was managed by an individual identified as 'Deadshot,' who developed personal dossiers on European Parliament members and journalists to identify pressure points within European political institutions.
What measures has Anthropic implemented to prevent future misuse of its AI tools?
In response to the findings of misuse, Anthropic has banned all accounts associated with these campaigns and implemented additional safeguards to prevent similar operations in the future.
Comments
Comments are moderated before publish.
No comments yet — be the first.