The Russia-linked operation, which OpenAI called “Dark Clark,” targeted countries across Latin America and focused heavily on undermining Ukraine’s reputation. Some activity also sought to influence domestic political debates, particularly in Argentina and Bolivia.
OpenAI said the operators appeared to control an organization called the Social Research Center, or SRC, while using a fabricated persona named “Mia Clark” to interact with people working for the group. Evidence reviewed by OpenAI indicated that employees in Latin America did not know they were working for a Russian operation.
The SRC produced research and maintained staff in the region, making the organization more elaborate than many of the influence networks OpenAI had previously disrupted. Internal reports generated with ChatGPT included references to hiring, firing, salaries, staffing and research projects.
The operators also reported efforts to distribute fabricated stories, documents and audio recordings. Some campaigns targeted Ukraine or attempted to weaken recruitment for the Ukrainian military, while others focused on political figures and events in Latin America.
OpenAI said some of the material appeared to spread far enough to trigger fact checks, media coverage and official denials. One campaign involving a fabricated message sent to schools in Peru later surfaced in reporting in Latin America and Europe and contributed to public tensions involving Ukraine and Poland.
ChatGPT was used most often by the Russian operators to prepare internal reports about their activities. OpenAI said the group sometimes claimed credit for developments it could not have caused, making its own descriptions of its influence unreliable.
The company rated Dark Clark as Category 5 on the IO Breakout Scale, which ranges from 1 to 6. OpenAI said it is the first operation at that level the company has disrupted since it began reporting on influence campaigns.
The second operation, which originated in Iran, was dubbed “Bogus Bylines.” It used seven fake journalist identities to pitch English-language articles to online publications, while also generating batches of social media comments in English and Persian.
OpenAI said the operation relied on virtual private networks to conceal where its users were located. Some of the fake journalist profiles claimed to represent Western writers, including American freelancers, even though account transparency information connected some of the associated social media accounts to Iran.
The Iranian operators used ChatGPT to review and refine articles, adapt pitches to publication requirements and prepare emails for editors. OpenAI identified almost 100 articles published or syndicated under the operation’s fake bylines across roughly a dozen online outlets.
Most of those articles focused on the conflict involving the United States and Iran. The operation also generated social media replies about the conflict and other political topics, sometimes producing comments that supported articles created under its own fabricated journalist identities.
OpenAI said those social media efforts generally produced little visible engagement. The published articles reached further because they appeared in established online outlets and could then be promoted through those publications’ own social accounts.
The company rated the article-publishing portion of Bogus Bylines as Category 4 on the IO Breakout Scale, while its social media commenting activity received a Category 2 assessment.
Both operations also used ChatGPT to prepare internal performance reports. OpenAI said the actors sometimes relied on misleading methods to make their campaigns appear more successful than available evidence supported.
The company said false-front operations can use AI to improve efficiency, language quality and editorial output while still relying on tactics that existed before generative AI. By creating convincing personas and organizations, operators can place material before audiences that may not know who produced it or why.
OpenAI said it disclosed information about both operations to relevant authorities and shared details from the Iranian case with industry partners. The company said publishing these investigations is intended to help researchers identify similar campaigns and make it more difficult for exposed operations to continue.
This analysis is based on reporting from The Hill & OpenAI.
Image courtesy of OpenAI.
This article was generated with AI assistance and reviewed for accuracy and quality.
