Iranian operators placed 12 ChatGPT-generated articles in US outlets 2026

Iranian operators created pseudonymous contributor accounts on US news platforms and fed topic prompts into ChatGPT-4o to produce draft articles on domestic policy and elections. Submissions bypassed initial editorial filters at three mid-tier outlets before publication. Microsoft Threat Intelligence tracking tied the originating IPs and account creation patterns to prior IRGC-linked personas documented in the 2024 Muddled Libra report.

Linguistic forensics showed elevated rates of repetitive syntactic structures and low variance in source attribution consistent with ChatGPT-4o temperature settings at 0.7. Cross-reference with Graphika’s 2025 AI Influence Operations dataset confirmed stylistic overlap with earlier Iranian test campaigns. No outlet applied watermark detection or multi-source verification on the accepted pieces.

Media intake processes treated the submissions as standard freelance contributions without requiring primary documentation or author credential checks. This gap allowed content to reach audiences before takedowns occurred. Outlets that implemented post-publication audits removed the articles within 72 hours after external notification.

Future submissions will likely incorporate local editing layers to reduce detectable artifacts. Outlets maintaining only single-stage review face continued exposure until credential and provenance checks become mandatory.