OpenAI blocks foreign influence operations

A moderation case with geopolitical stakes

OpenAI’s latest threat report turns a familiar warning about generative AI into a concrete enforcement case: foreign influence operators are not merely experimenting with mainstream AI tools; they are folding them into old-school propaganda workflows. On October 8, the company said it had banned two covert influence operations, one originating in Russia and one in Iran, that used ChatGPT alongside traditional tradecraft to launder geopolitical messages through seemingly independent organizations and authors [1].

The Russian operation, dubbed “Dark Clark” by OpenAI, used ChatGPT in support of campaigns across Latin America, where its apparent goals included undermining Ukraine’s reputation and influencing domestic politics in countries such as Argentina and Bolivia [1]. The Iran-origin operation, called “Bogus Bylines,” used seven fake journalist personas to pitch and polish long-form articles about the U.S.-Iran conflict for online outlets around the world [1]. In both cases, OpenAI’s enforcement action was less about a single viral deepfake than about the discovery of repeatable operating models: fake institutions, plausible identities, editorial polishing, internal reporting and attempts to measure or exaggerate impact.

Russia’s “think tank” front

The Russian case stands out because it allegedly blended AI assistance with human infrastructure on the ground. OpenAI said the operators appeared to control a Latin America-facing “research platform” called the Social Research Center, or SRC, through a fabricated persona named “Mia Clark” [1]. The company said the available evidence indicated that the SRC’s local employees in Latin America did not know they were working for a Russian group, and that the platform produced a majority of its material through those co-opted staff rather than purely through AI generation [1].

That detail matters. It suggests a shift from the caricature of the influence bot farm toward a more layered model: hire or recruit real people, place them behind a benign-looking institution, then use AI to make the back office faster and smoother. According to OpenAI, the operators used ChatGPT mainly to draft internal reports to superiors, summarize activity and prepare assessments, though they also used it in a smaller number of cases to create fake materials, including leaked documents and audio scripts [1]. Recorded Future News likewise reported that Russian operatives used ChatGPT to help run a covert campaign in Latin America and that some false claims spread far enough to trigger fact-checks or official denials [3].

OpenAI assessed the Russian network as Category 5 on the Breakout Scale, a six-level framework for influence operations, saying it was the first Category 5 operation the company had disrupted since it began public reporting on such campaigns [1]. The company pointed to examples in which claims linked to the operation appeared to feed public tensions, including a fake school-related incident in Peru that later surfaced in Latin American and European media and prompted comment from a Polish member of the European Parliament [1]. NPR’s report emphasized the same pattern: Russian operators used AI not only to draft fake material but also, in many cases, to update superiors and take credit for influence activity [2].

Iran’s “Bogus Bylines”

The Iranian operation followed a different playbook. Instead of centering on a local front organization, it relied on fake author identities that posed as Western journalists. OpenAI said the operation used seven bylines — including Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams and Alice Johnson — to pitch articles to editors [1]. The campaign’s most successful activity was not social media spam but the placement or syndication of nearly 100 articles across roughly a dozen small and medium online outlets focused on geopolitics, international affairs and the Middle East [1].

The use of media submissions is crucial because it gives influence operators borrowed credibility. A post from a throwaway account can be ignored; a bylined article on a real website can be shared, cited and promoted as if it came from an ordinary contributor. OpenAI said some of the personas were supported by social media accounts on platforms including X, Facebook, Instagram, Threads and Medium, while transparency information for one persona’s accounts pointed back to Iran [1]. AFP and Euronews both reported that the Iranian campaign planted long-form articles under fake Western-seeming journalist profiles and generated social media comments about the U.S.-Iran war [4], [5].

The Iranian network also used ChatGPT to refine English-language articles, check drafts against outlet submission criteria and generate pitch emails to editors [1]. The campaign’s comments on social platforms, by contrast, appear to have had limited traction. OpenAI said the replies it could identify generally showed low engagement, while the article-placement effort reached a higher Category 4 assessment because some outlets had substantial followings [1]. NPR reported that Daily Kos and Middle East Monitor were among the outlets where articles by fake personas appeared, and quoted Middle East Monitor as saying it was reviewing submissions and strengthening editorial safeguards [2].

AI made the old tactics easier, not magically new

The key lesson is not that ChatGPT invented foreign propaganda. It did not. The tactics OpenAI describes — fake journalists, front organizations, narrative laundering, forged leaks and inflated internal claims — long predate generative AI. OpenAI itself compared the cases to older influence operations that used deceptive journalist or media fronts, while noting that AI made some workflows easier [1].

That distinction should shape the policy response. If the problem is treated as a purely technical failure, the answer will be too narrow: better classifiers, stricter prompts, faster bans. Those tools matter, but the operations described here exploited human trust as much as model output. Editors accepted pitches. Local workers contributed to a front institution. Audiences encountered narratives through familiar publication formats. The vulnerability was partly linguistic fluency and scale, but also provenance: who is behind this author, this research center, this “leak,” this sudden wave of comments?

For AI companies, the case raises expectations around attribution, account controls and transparency reporting. OpenAI said the Russian operators used VPNs to reach its services from Russia, where it does not allow access, and the Iranian operators also obscured their location through VPNs [1]. That underscores a recurring enforcement dilemma: access restrictions are only as strong as detection, identity signals and cross-platform intelligence sharing. OpenAI said it shared information from these cases with relevant authorities and industry partners [1].

What comes next for platforms and newsrooms

For the wider industry, the episode is a reminder that general-purpose assistants can become geopolitical force multipliers when paired with deceptive distribution. The most damaging use may not be fully automated propaganda at massive scale; it may be a more mundane productivity gain for influence teams that need cleaner English, localized Spanish, plausible bios, polished pitches and tidy reports for their managers.

News organizations also have a role to play. The Iranian case shows that community sections, opinion submissions and small international-affairs outlets can become targets because they offer a route into the legitimate information ecosystem. Stronger contributor verification, clearer labeling of community content, checks for recycled biographies and better escalation channels with platforms could reduce the attack surface without closing the door to genuine outside contributors.

OpenAI’s bans will not end state-linked influence work, and the company’s report does not prove that every piece of content in these operations was AI-generated. But the disruption makes one thing clear: the AI influence threat is no longer hypothetical. It is operational, hybrid and bureaucratic. Skynet, as it turns out, may not begin with killer robots; it may begin with a content-moderation queue, a VPN login and a very polished pitch email.

Sources from the last 72 hours

1. [1]Disrupting AI-enabled “false front” operationsOct 8, 2026, 2:00 AM
2. [2]OpenAI says it disrupted Russian and Iranian influence ops that used ChatGPTOct 9, 2026, 2:00 AM
3. [3]OpenAI says Iran, Russia influence ops bustedOct 9, 2026, 3:24 AM
4. [4]Iran and Russia used ChatGPT for propaganda by ‘fake journalists’, OpenAI saysOct 9, 2026, 1:53 PM
5. [5]OpenAI caught Russians and Iranians using ChatGPT for influence campaignsOct 8, 2026, 2:00 AM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.