OpenAI says Iranian operatives used ChatGPT to plant critical news articles

How the Bogus Bylines Operation Worked

According to OpenAI, the operators worked with its models mostly in Persian. They used them for three jobs: drafting long, polished articles, writing editorial pitches, and churning out social media comments.

The Fabricated Personas

The seven fabricated personas carried Western-sounding names, including Michael Harrison and Ervin B. Hoskins. Under those bylines, the campaign placed approximately 100 articles across about a dozen outlets.

The Scale of the Operation

One headline OpenAI cited read 'The Iran War and the Collapse of Western Moral Credibility.' The operation ramped up after the US-Iran conflict escalated, with output peaking during the fighting.

The Uncertain Affiliations

OpenAI rates influence operations on what it calls the IO Breakout Scale. Bogus Bylines landed at Category 4. A Russian effort the company also disrupted, named Dark Clark, received a rating of Category 5, a higher tier indicating a more serious threat.

The Implications for Media and Readers

Bogus Bylines exploited the contributor submission system. If seven invented writers could place roughly 100 articles across about a dozen outlets, contributor vetting at many publications is thinner than readers might assume.

The Future of AI-Generated Content

What this means for media, platforms, and readers is that AI-generated content is becoming increasingly sophisticated. OpenAI's models helped produce the content, but OpenAI also spotted the activity, shut down the accounts, and published the details.