What influence operations are

AI-generated personas in influence operations

Q: Has AI made influence operations more effective?

Meta said in December 2024 gains were only incremental. OpenAI said in May 2024 its services did not bring a meaningful increase in engagement.

Q: How many EU-recorded incidents used AI in 2025?

The EEAS counted 147 of 540 incidents, or 27 percent.

Q: Which networks used AI video presenters?

Meta reported AI newsreaders in a Lebanon-origin network, and OpenAI reported one in Stop News.

Last checked 24 October 2026 · 10 sources

Short answer

Takedown reports describe AI use in three main forms: generated profile photos, AI-written posts and comments, and synthetic video presenters. Meta has reported GAN profile photos in many networks and AI newsreader videos in a Lebanon-origin network ³. OpenAI said in May 2024 that operations used AI to post more text with fewer language errors ⁶. The EEAS counted AI-related techniques in 27 percent of 540 incidents in 2025, up from about 41 cases in its previous report ^10,9.

EEAS, 2025: 147 cases with AI-related techniques, 27 percent of 540 incidents ¹⁰
EEAS, Nov 2023 to Nov 2024: About 41 cases of AI use in 505 incidents ⁹
Meta view, December 2024: GenAI gave threat actors only incremental productivity and content gains ³
OpenAI view, May 2024: None of five operations used AI exclusively ⁶

What counts as an AI-generated persona

Reports use the term for a fake identity whose face, words or voice were made with AI tools. The most common form in Meta reports is a profile photo likely created with a generative adversarial network, or GAN. Meta reported GAN photos in a Croatia-origin network and in an unknown-origin network targeting Moldova and Madagascar in May 2024 ¹.

Meta's Q4 2024 report said the Benin-based network used profile photos likely generated using AI. It said OpenAI reported that comments in the Ghana network were created with GenAI ⁴.

Synthetic newsreaders and journalists

Meta said a Lebanon-origin brand called Israel in a Minute frequently posted AI-generated video newsreaders. The same network used one fictitious journalist persona with a GAN profile photo ³. In August 2024 Meta reported AI-generated news-reader videos on YouTube from a Russia-origin network aimed at English and French speakers ².

OpenAI said in October 2025 that its Russia-origin case Stop News used an AI-generated newsreader on YouTube and TikTok whose appearance changed through 2025. OpenAI did not confirm which service generated it ⁸.

A Meta indicator file for its H1 2026 report describes a Pakistan-based network that Meta attributed to the military's Inter-Services Public Relations wing. Meta said it used AI video generation to create photorealistic journalist personas for video interviews, and large language models to write posts in English and Urdu ⁵.

AI-written text and fake engagement

OpenAI's first public report on 30 May 2024 covered five operations. It said all of them used AI to some degree but none used it exclusively ⁶. It said some networks used its services to help create the appearance of engagement ⁶.

In June 2025 OpenAI described a China-origin case, Uncle Spam, that used AI-generated profile images and logos for personas posing as US veterans ⁷. It described a Russia-origin case, Helgoland Bite, that used an AI-generated profile image for an X account during the 2025 German election ⁷.

Meta said in May 2024 that some comments by the STOIC network were likely generated using AI ¹.

How reporting bodies assess the effect

Meta said in May 2024 that it had not seen novel GenAI tactics that would impede its ability to disrupt networks ¹. In December 2024 it said GenAI had given only incremental productivity and content gains ³.

OpenAI said in October 2025 that actors were building AI into existing workflows rather than building new workflows around AI ⁸.

The EEAS counted about 41 AI cases between 4 November 2023 and 4 November 2024 ⁹. Its 4th report counted 147 cases in 2025, 27 percent of incidents ¹⁰. The EEAS says its data is a selective sample ¹⁰.

Common questions

How do platforms spot AI-generated profile photos?

Meta reports describe photos as likely created using GANs, an assessment made during its investigations ^1,3.

Has AI made influence operations more effective?

Meta said in December 2024 gains were only incremental ³. OpenAI said in May 2024 its services did not bring a meaningful increase in engagement ⁶.

How many EU-recorded incidents used AI in 2025?

The EEAS counted 147 of 540 incidents, or 27 percent ¹⁰.

Which networks used AI video presenters?

Meta reported AI newsreaders in a Lebanon-origin network ³, and OpenAI reported one in Stop News ⁸.

Rules and charges change. Check the official source linked on this page before you pay or travel. Spotted an error? Email hello@influenceopsrecords.com.

Sources

Influence operation tactics documented in official reports
Fake personas, fake media brands, paid ads, self-amplification and AI-written content, as Meta, Google, OpenAI and the EEAS have documented them.
OpenAI threat reports explained
What OpenAI's reports on disrupting malicious uses of AI contain, how cases are rated on the Breakout Scale, and what the 2024 to 2026 reports found.
Fake news websites in influence operations
Spoofed outlets, invented news brands and aggregator networks as documented by Meta, Microsoft, OpenAI, Google and the EEAS, with dates and counts.
How platforms detect fake networks
What Meta, OpenAI, TikTok and Google say about finding influence networks: expert investigations, automated detection, behavioural signals and shared tips.