OpenAI bans Russian network Dark Clark targeting Latin America

OpenAI announced on October 8, 2026, that it blocked ChatGPT accounts used by an influence operation it named Dark Clark, which it attributes to actors in Russia. The campaign focused its activity on Latin America and is the first operation dismantled by OpenAI to receive a category 5 rating on a six-level scale.

The framework used is the Breakout Scale, published by Ben Nimmo at Brookings in September 2020 to measure cross-platform reach rather than persuasion. Under this model, category 4 represents amplification by news media, category 5 reflects amplification by high-profile figures such as politicians or celebrities, and level 6 covers political action or calls to violence.

OpenAI linked the operation's records to an incident in Ecuador, where a June 17, 2026, fact-check by Ecuador Chequea confirmed a real school event where students swore loyalty to President Daniel Noboa and Erik Prince. Minister of Education Gilda Alcivar denied ordering the ceremony, and the media outlet found irregularities including a fake Gmail sender, an invalid circular number format, and an unauthorized signature on documents from April. OpenAI noted that the operators used ChatGPT primarily to write internal reports.

OpenAI also tied some accounts to Politology, an entity also known as La Compañía. An earlier investigation published on February 14, 2026, by Forbidden Stories reviewed 76 documents spanning 1,431 pages from 2023 and 2024 operations, noting that the network used local intermediaries and exaggerated results in internal briefings. The Russian intelligence service and network members did not respond to inquiries at the time, and no response was registered to OpenAI's findings.