Global Security Feature: Information Operations, 2026-W41

Bottom line #

On 2026-10-08 OpenAI disclosed that it had shut down two covert, AI-assisted influence operations: a Russian one aimed at Latin America, known as "Dark Clark", and an Iranian cluster built on fake journalist identities. OpenAI rated them 5 and 4 on its 1-to-6 impact scale; it says few campaigns rise above 1 or 2. In the same week StopFake republished DFRLab's finding (2026-10-06) that a fabricated "Politico" claim about Armenian tomatoes was pushed by coordinated actors into eighteen languages in 29 hours, and Montenegro ordered the head of the pro-Kremlin Rybar network out of the country.

This week's documentation covers only Russia-linked and Iran-linked actors. No operation by the US, Canada, an EU member, Ukraine or Israel was documented. The attributors are mostly Western or Ukrainian bodies, so the imbalance likely reflects who funds and publishes investigations as well as who runs operations (moderate confidence: the pattern is consistent, but no source measures it).

Cases #

Dark Clark: Russian front-identity operation in Latin America (IO-2026-W41-01) #

• Actor: Russian actors, per OpenAI. The report does not attribute the activity to any specific Russian government or intelligence agency, but OpenAI said evidence suggested connections to a Russian organization known as Politology or La Compania. Channel class: state-aligned (no tie to a state body documented).
• Behaviour: A fake AI persona, "Mia Clark", fronted a supposed Latin American think tank, the Social Research Center. Co-opted staff produced most of the original content. Operators used ChatGPT to write fake leaked documents and scripts for audio recordings, and reached OpenAI's services through VPNs.
• Content: Mainly harming Ukraine's reputation in Latin America, plus political issues in Argentina and Bolivia. Examples: a TikTok video falsely claiming Ecuador used shell companies to recruit citizens to fight for Ukraine, and a fake audio clip in which a state water company employee claimed La Paz was about to lose its water supply.
• Degree: Breakout 5 on OpenAI's scale. The content drew fact-checks and an official government denial.
• Effect: An official denial from the Bolivian government (the La Paz audio, during protests in May); a fact-check by Lupa Media (the Ecuador video, in March).
• Four-criteria test: met (false content, deliberate spread through front identities, harm to state interests, a foreign actor). This is an operation.
• Kill-chain phases described: acquiring assets, disguising assets, evading detection, targeted engagement.
• Status of core claims: false, per OpenAI and the fact-checks and government denial it cites.
• Sources: CyberScoop; The Record.
• Attribution: OpenAI, 2026-10-08; grade C:p O:y E:p T:p; not corroborated.

Iranian fake-journalist article placement (IO-2026-W41-02) #

• Actor: An Iranian cluster of fake journalist identities, per OpenAI. The operators mostly wrote Persian-language prompts and used VPNs. OpenAI said the activity resembled "a commercial actor running a for-hire influence campaign" and did not attribute it to a specific actor or group. Channel class: undetermined.
• Behaviour: At least seven fake journalist personas pitched articles to editors. ChatGPT was used to write the articles and the pitches.
• Content: Articles on the US-Iran war.
• Degree: Almost 100 articles published or syndicated across roughly a dozen small and medium outlets. One of them had nearly 2 million Facebook followers. The stories drew few comments or social media reactions. Breakout 4 on OpenAI's scale.
• Effect: None observed beyond publication.
• Four-criteria test: not met. Deceptive spread and a foreign actor are documented; no source says the articles were false or harmful. This is covert influence activity, not an operation under the test.
• Kill-chain phases described: acquiring assets, disguising assets, evading detection, targeted engagement.
• Status of core claim: unverified (the articles' accuracy has not been assessed).
• Sources: CyberScoop; The Record.
• Attribution: OpenAI, 2026-10-08; grade C:p O:y E:p T:p; not corroborated.

Fabricated "Politico" claim on Armenian tomatoes (IO-2026-W41-03) #

• Actor: A pro-Kremlin amplification network, per DFRLab: the Pravda Network's mirror sites and X accounts with a history of spreading Storm-1516 content. The biggest single amplifier was a prominent pro-Kremlin Telegram blogger. No state body is named. Channel class: state-aligned.
• Behaviour: Coordinated amplification involving inauthentic accounts. DFRLab compiled over 380 original posts. Of the X accounts involved, 74 had a documented history of reposting or quoting content from campaigns DFRLab attributes to Storm-1516.
• Content: A claim that Politico reported the EU had rejected Armenian tomatoes. Politico published no such report.
• Degree: Eighteen languages in 29 hours across four platforms (Telegram, X, Facebook and VK). The first post was on 2026-07-22. Breakout 3.
• Effect: None observed.
• Four-criteria test: partly met. False content and coordinated spread are documented; harm to a state's interests and the actor's tie to a state are only partly documented. Recorded as a narrative push with documented coordination.
• Kill-chain phases described: disguising assets, indiscriminate engagement.
• Status of core claim: false, per DFRLab.
• Sources: DFRLab; StopFake.
• Attribution: DFRLab, 2026-09-29; grade C:p O:y E:y T:p; not corroborated.

Rybar network head ordered out of Montenegro (IO-2026-W41-04) #

• Actor: Rybar, a pro-Kremlin military information network. The UK government has identified Mikhail Zvinchuk as Rybar's co-owner and CEO, and says the network is partly coordinated by the Russian Presidential Administration, funded by Rostec and works with Russian secret services. The US State Department lists him as involved in influence operations targeting US elections. Channel class: state-linked.
• Behaviour: Vijesti, citing security sources, reported that authorities suspected the group of planning a documentary. Police seized drones, high-resolution recording equipment, cameras and mobile phones. The five had arrived less than two days earlier.
• Content (suspected, unpublished): A documentary meant to undermine support for Montenegro's European integration, portraying its alignment with EU visa policy as an imposed measure that threatens tourism and the economy.
• Degree: No content in circulation is documented.
• Effect: Montenegro ordered Zvinchuk and four other Russian citizens deported on national and internal security grounds. Former Russian President Dmitry Medvedev suggested Moscow might need more prisoners to exchange to secure Zvinchuk's return.
• Four-criteria test: not met. No false content is in circulation and the harm is only suspected. Recorded as a state action against a suspected operation.
• Status of core claim: unverified. The documentary plan rests on anonymous security sources.
• Sources: Euromaidan Press; Ukrainska Pravda; Ukrainska Pravda.
• Attribution: UK government and US State Department, dates not stated in the reports; grade C:p O:p E:n T:n; corroborated by Montenegro's police action.

Kremlin narratives on Bucha and Ukraine, per EUvsDisinfo (IO-2026-W41-05) #

• Actor: Russian state-controlled outlets (Sputnik Georgia; an executive of the VGTRK broadcasting group) and The Other Ukraine, which EUvsDisinfo calls a Russian FIMI outlet. Channel classes: state-controlled and state-aligned. Statements by officials themselves are open messaging and outside this feature.
• Behaviour: Recycled themes. EUvsDisinfo says Bucha is among the topics most heavily targeted by coordinated pro-Kremlin amplification on social media.
• Content: That the Bucha massacre was staged by Ukraine and the West; that Ukraine hides a child-trafficking business; that Russian strikes are part of a "fight against Nazism" in Europe.
• Degree: Carried by state-controlled media. Breakout 4.
• Effect: None observed.
• Four-criteria test: a recurring narrative push; no account or spread data is given.
• Kill-chain phases described: indiscriminate engagement.
• Status of core claim: false, per EUvsDisinfo.
• Sources: EUvsDisinfo; EUvsDisinfo.
• Attribution: EUvsDisinfo (EEAS East StratCom), 2026-10-05 and 2026-10-09; grade C:p O:p E:p T:p; not corroborated.

Matryoshka fake news-brand reports (IO-2026-W41-06) #

• Actor: Matryoshka, which NewsGuard calls a Russian disinformation campaign. Channel class: state-aligned.
• Behaviour: Fake news reports imitating credible media brands and institutions.
• Content: 20 fabricated reports on the French student protests in two days, and fabricated video reports claiming the US Democratic Party orchestrated the FlyDubai incident.
• Degree: NewsGuard says the protest reports drew little traffic. Breakout 1.
• Effect: None observed.
• Four-criteria test: harm to a state's interests is not documented.
• Kill-chain phases described: disguising assets.
• Status of core claim: false, per NewsGuard.
• Sources: NewsGuard Reality Check; NewsGuard Reality Check.
• Attribution: NewsGuard, 2026-10-07 and 2026-10-08; grade C:p O:p E:p T:p; not corroborated.

Noise floor #

FlyDubai. What is known: Israeli, Emirati and Saudi authorities are investigating a foiled hijacking of a FlyDubai flight in which a co-pilot allegedly attacked the captain (NewsGuard). The claim: the incident was a "false flag", and Fars said Israel had considered attacking the aircraft to blame Iran. Why it fails: NewsGuard found no evidence for either version; the investigations will settle the facts.

Attributions this week #

The record #

This week: 6 case(s) and 7 unverified claim(s) logged, 6 of them described in the Noise floor. Since the record began: 6 cases and 8 claims.

Cases this week by actor state

Everything logged this week by channel class

Status of the core claim

Breakout category (observed reach, 1 to 6)

By week

Counts include every item assessed this week; minor claims are counted but not described.

Context and baseline #

• Against 2026-W40: the previous week had no documented case and one checked claim, a doctored 2020 clip of Ethiopian Prime Minister Abiy Ahmed. This week has six cases, one of which (Dark Clark) meets all four criteria for an operation. Nothing Ethiopia-related recurred.
• Recurring cases: none on record yet.
• Scale: OpenAI says few of the campaigns it rates rise above 1 or 2 on its 1-to-6 scale, so ratings of 5 and 4 are high. It called Dark Clark "the most complex attempt to run a front identity that we've disrupted over the past two and a half years."
• Pattern, not attribution: Ukraine-focused fact-checks this week cluster on themes of allies abandoning Kyiv (grain transit, aircraft deliveries) and of internal division (MPs leaving the capital, language categories). This is an observed pattern, not an attribution; no source links these items to one actor (low confidence that they share an origin).
• Imbalance: all six cases involve Russia-linked or Iran-linked actors.

Indicators to watch #

• OpenAI or other AI companies naming the outlets that carried the Iranian-persona articles, or tying Politology to further operations.
• Whether Montenegrin authorities publish evidence on the Rybar group's planned content, and any Russian response beyond Medvedev's remark.
• Matryoshka fabrications on French or other EU events reaching beyond one platform (Breakout 2 or higher).
• New "false flag" narratives after major incidents, and state outlets amplifying them.
• Further DFRLab or EEAS reports on the Pravda Network or Storm-1516 accounts.
• Any operation documented as run by a Western state, Ukraine or Israel, graded on the same terms.

How this was made #

This week drew on 15 sources: three investigator and government-body analyses (DFRLab, two from EUvsDisinfo), six fact-checks (two from NewsGuard, four from StopFake) and six news reports (CyberScoop, The Record, Euromaidan Press, two from Ukrainska Pravda, Kyiv Independent). OpenAI's own report, the UK and US government listings on Rybar, and Vijesti's original report could not be checked directly; those details rest on the news coverage cited. The method is on the methodology page.

How this was made: methodology and limits.