JA EN
shiichan

A First Category 5! OpenAI Disrupts Two AI-Enabled "False Front" Influence Operations

OpenAI News ★★★ View Markdown
#openai#ai-safety#security

Hi everyone, Shii-chan here! Today I read a report about stopping the misuse of AI. It's a serious topic, but I'll walk you through it carefully!

OpenAI News openai.com

What was announced?

In OpenAI's News, OpenAI reported disrupting two AI-enabled "false front" influence operations (IO), one from Russia and one from Iran. They used OpenAI's models together with more traditional techniques and technologies.

Both used fake entities to launder conflict-related messaging into their target audiences:

  • Iran-origin "Bogus Bylines": seven "journalist" personas pitched long-form articles to small and medium online outlets around the world
  • Russia-origin "Dark Clark": apparently co-opted unwitting people in Latin America to run a "think tank" on the ground

Why it matters

OpenAI has been publishing these reports for about two and a half years. The goal is to inform regulators, industry peers, and society about how threat actors try to use AI.

What stands out this time is the potential reach. On the Breakout Scale, which rates influence operations from 1 (lowest) to 6 (highest):

  • The Russia-origin operation was assessed as Category 5, the first Category 5 operation OpenAI has disrupted since it began reporting
  • The Iran-origin operation reached Category 4

Both landed content in mainstream media outlets, not only on social media.

What changes

The aim of reporting is to make further research and disruption easier, and to make continuing such operations harder. OpenAI notes that "Alice Donovan" and "PeaceData" both ceased activity after being exposed.

For editors, media professionals, and platform investigators, this is a concrete look at the tactics involved.

Dive Deep

Iran-origin "Bogus Bylines"

  • Prompts were in Persian, output in Persian and English, and the actors used VPNs to hide their location
  • They asked the model to check English drafts against a specific outlet's submission criteria, then generate the pitch email to the editor
  • Seven fake bylines were used, and some were backed by fake accounts on social platforms
  • Open-source investigation found almost 100 articles under these bylines across over a dozen publications. The earliest was in July 2025, and the pace rose sharply after the US-Iran conflict began
  • They also generated batches of social media comments, portraying the US as the aggressor, Israel as an unreliable ally, and Iran as a defiant but resilient victim
  • They measured impact by the views of the posts they replied to, which greatly exaggerated their effectiveness

Russia-origin "Dark Clark"

  • Access to OpenAI's models is not allowed from Russia, so the operators used VPNs
  • Main uses were internal reports to a superior, content for the operation, and performance reports on a self-described "research platform" called the Social Research Center (SRC)
  • They appear to have controlled the SRC via a fake persona, "Mia Clark", which gave the operation its name
  • Goals were to undermine Ukraine's reputation and to influence local politics, especially in Argentina and Bolivia
  • The SRC's staff in Latin America appear not to have known they worked for a Russian group. OpenAI calls it the most complex attempt to run a front identity it has disrupted in the past two and a half years
  • Internal reports described things like 2024 campaigns about Argentina's president and a fake email to schools in Peru in May 2026
  • Some fakes spread widely enough to prompt fact checks and official denials

OpenAI's view is that these operations closely resembled complex pre-AI influence operations, with AI making some workflows easier.

Wrap-up

  • OpenAI disrupted Russia-origin and Iran-origin "false front" influence operations
  • The Russian one was rated Category 5 (a first), the Iranian one Category 4
  • They used fake journalists and a "think tank", and landed content in mainstream media
  • AI streamlined polishing text, pitching, and internal reporting
  • Exposure tends to stop such operations, so reporting is also a way to share information for defense

This is a good read for media editors, trust and safety teams, and anyone interested in preventing AI misuse!

Related posts