#

Key takeaways

• OpenAI says it disabled two covert influence operations.

• One operation originated in Russia and targeted audiences in Latin America.

• The other originated in Iran and used invented journalist personas.

• OpenAI says AI helped draft internal reports, but those reports may have overstated results.

• The report describes observed tactics and enforcement, not independent proof of every claimed effect.

OpenAI's October 8 security report

OpenAI says it disabled two covert influence operations that used ChatGPT alongside older methods. The company describes both efforts as attempts to create apparently independent fronts. It presents the findings as an enforcement and assessment report.

The October 8 report describes two influence operations that OpenAI says it disrupted. Its account distinguishes company observations from assertions made by the operators. That matters because the report is about observed tactics, not a full verification of every claimed outcome.

What OpenAI says it found

According to OpenAI, one operation originated in Russia and targeted audiences in Latin America. OpenAI says the effort appeared to direct a research organization in Latin America. It also says local staff did not know who controlled it.

OpenAI says the other operation originated in Iran. It says the Iranian effort used seven invented journalist personas. Those personas were used to pitch long articles to smaller online publishers. OpenAI also says the effort generated groups of social-media comments.

The report says the operators used AI heavily to draft internal reports. OpenAI says those reports sometimes overstated results. The source also says some material reached mainstream outlets, but OpenAI cautions that not every published item was made with its models.

How OpenAI frames the scale and evidence

OpenAI assigns the Russia-origin activity to category five on the cited influence-operation reach scale. It assigns the Iran-origin activity to category four. OpenAI describes the cited Breakout Scale as ranging from one, the lowest potential reach, to six, the highest. These ratings are OpenAI assessments of potential reach.

The report also describes fake documents and scripts, along with ordinary editorial placement attempts. That mix suggests the operations used both fabricated materials and standard outreach methods. Still, the source is careful to separate OpenAI's assessment from independent verification.

Why the distinction matters

The source repeatedly warns against treating operator claims as proven facts. OpenAI's report is about what it observed and what it disabled. It is not proof that every claimed result happened exactly as described.

That distinction is important for readers. A covert influence operation can produce internal reports, pitches, and content drafts without proving reach or impact on its own. OpenAI's own caution shows that published outcomes and operator claims are not the same thing.

Morocco relevance

The source reports no Morocco-specific target, publisher, operation, or policy response. The only safe lesson is conditional and global: readers should separate platform assessments from independent verification when reviewing influence-operation claims.

What readers should watch for

This report highlights a familiar pattern. Operators may combine AI-generated text with older tactics such as fake identities, editorial pitching, and fabricated documents. That combination can make a front look more legitimate than it is.

OpenAI compares these cases with 30 influence operations it has exposed over roughly two and a half years, while assessing the reach of the two newly described operations separately. It also does not claim that every item linked to the operations was AI-generated. Readers should therefore treat the report as a documented enforcement action, not a blanket statement about all related content.

Bottom line

OpenAI says it disrupted two covert influence operations and published its assessment on October 8. One case originated in Russia, and the other originated in Iran. Both used AI in support of broader deception efforts.

The main takeaway is simple. AI can speed up influence work, but it can also leave traces that platforms can detect. The report shows OpenAI acting on those traces, while also warning readers not to overread the operators' own claims.