How the Kremlin is using new digital tools to reshape propaganda
You no longer need real people to create propaganda. Now, AI can do it all. Fake people, fake images, fake stories—all driving the messages of your choosing. In this investigation, Tatsiana "Tanya" Ashurkevich and Christo Grozev go inside the market that manufactures Russia's AI fakes. And they buy their own.
This video was produced in partnership with the Pulitzer Center's AI Accountability Network. Working from a collection supplied by Bot Blocker, an anonymous research collective that tracks Russian troll and bot networks, we looked at how the Russian AI fakes are actually built. There are two different techniques.
The first technique steals a real person, usually a celebrity. Their real footage is kept, their voice is cloned by AI, and the middle of the clip is replaced with whatever the Kremlin might like to see. Some of the celebrity videos begin with real Cameos, the website where anyone can pay a famous person for a short personalized greeting. Then AI is leveraged to make the celebrity do and say things they never did or said.
The second technique invents a person outright. No filming and no actors. With AI, a persona can appear in any language, any accent, any biography and then the message and messenger can be tailored to a specific audience.
Once establishing these tactics, we then decided to test the whole pipeline ourselves.
Ashurkevich opened an account on a Russian freelance marketplace under a made-up name and ordered a fake: Grozev and her on a yacht with a NATO general, shot to look like paparazzi footage. Nobody asked who she was or what the video was for. The AI model refused the job once, telling the freelancer that a real recognizable person placed in a staged scene is "exactly how fake compromising photos are built for disinformation." The freelancer worked around the refusal anyway, and while he was building it he wrote: "I just hope they don't recognize Christo Grozev."
About an hour later the fake was ready, for roughly $80. Finding someone to publish it took ten minutes and another $80.
As a nonprofit journalism organization, we depend on your support to fund more than 170 reporting projects every year on critical global and local issues. Donate any amount today to become a Pulitzer Center Champion and receive exclusive benefits!
Within an hour, six Russian news sites had published it as fact, under the headline: "Grozev spotted on a yacht with a NATO general and a red-haired girl." The general does not exist. The afternoon on the yacht never happened.
Why this matters:
Russia is not the only state that does this, but its operations are among the most advanced.
So we asked the people who are supposed to stop it. Google's chief threat analyst told us that who bears responsibility is "a policy question." OpenAI described how it banned a Russian network it calls Bad Grammar, which exposed itself when one of its bots published the model's own refusal on Telegram instead of the comment it was meant to post. It then confirmed the issue: that when one company shuts an operation down, it can simply reappear on another company's platform.
The European Commission declined an interview and twice avoided answering a question on whether AI companies are doing enough. Anthropic did not respond to repeated requests.
