Hosts on Venezuelan state-owned television station VTV have been touting positive news coverage about their country from “una agencia gringa” — an American news agency. “This information isn’t coming from VTV, it’s not coming from me … these are numbers from an American news outlet,” one host exclaimed while showing clips of English-speaking anchors reporting favorably on Venezuela hosting baseball’s Caribbean Series and the country’s tourism industry.
Deepfakes are Venezuela’s latest disinformation tool, experts say
But the reporters in those videos aren’t real. Their names are Daren and Noah, and they’re computer-generated avatars crafted by Synthesia, a London-based artificial intelligence company.
The clips are from a YouTube channel called House of News, which presents itself as an English-language media outlet. Researchers say the videos are part of the Venezuelan government’s attempts to spin the narrative on social media, considered one of the last bastions of free speech in a nation where outlets are censored and journalists are often persecuted. The incorporation of AI, experts told The Washington Post, seems to be a new addition to the government’s disinformation campaigns, which range from incentivizing Twitter users to post specific talking points to using bots that spit out the regime’s messaging.
On buses and balconies, Venezuela’s citizen reporters take news to the people
As companies compete to bring the technology forward, the episode in Venezuela is an example of AI’s “dark side,” said Hany Farid, a University of California at Berkeley professor and deepfakes expert.
“Not only are we developing these technologies, we’re also putting them on the internet, letting anybody use them and then sitting back and being like, ‘Well, let’s see what happens next,’” Farid said.
The user behind House of News has been banned by Synthesia for violating the terms of service, a spokesperson said, adding that the company apologizes “for any misuse of our platform.”
“It pains us to see people misuse the product we built to help benefit society — this was never our intention,” the company said in a statement. “However, we won’t let the minority ruin the good AI has to offer.”
Odd grammar and talking points
The videos began popping up in early February, said Adrián González, director of Cazadores de Fake News, a nonprofit that tracks disinformation in Venezuela.
Though the avatars look human, their speech lags, and they use odd grammar. The videos’ themes, González said, coincide “with the government’s efforts to put out this narrative to the world that Venezuela se arregló,” or the nation’s problems — including its ongoing immigration, political and economic crises — have been fixed.
Anyone with an iPhone can now make deepfakes. We aren’t ready for what happens next.
In one House of News video, a female avatar breaks down how Juan Guaidó, recognized internationally as Venezuela’s interim president until last year, allegedly wasted $150 million. The channel’s latest clip says Venezuela is one of the most visited destinations in Latin America, echoing President Nicolás Maduro’s recent claim that tourism is the “secret weapon” to “economic rebirth.”
“The Venezuelan government has a long history of disseminating misinformation, and it couldn’t be a coincidence that all these videos were suddenly pushing officials’ main talking points,” González said.
The videos, which have nearly 716,000 views, have been pushed as ads on YouTube — a frequent complaint in the comments. Users have also criticized the clips as “out of touch.”
It’s unclear who’s behind House of News. Its YouTube channel, which was opened on Jan. 26 and has five videos featuring Synthesia avatars, does not list a website or contact information.
Cazadores de Fake News determined the videos were deepfakes by running screenshots through PimEyes, a facial recognition tool — finding the avatars had been used in dozens of business and educational videos.
Most clients use the technology to create corporate training videos and real estate tours, a Synthesia spokesperson said. But this isn’t the first time Synthesia’s avatars have been used in disinformation campaigns. Last month, Graphika, a New York-based disinformation research firm, discovered a pro-China political operation with a similar modus operandi.
Tech’s hottest new job: AI whisperer. No coding required.
Synthesia, which also banned the user in that campaign, said the situations “highlight how difficult moderation is.”
“We founded the company in 2017 on strong ethical principles and have taken a much stricter stance than most,” the spokesperson said.
Last week, Maduro and Minister for Communication and Information Freddy Ñáñez denied using AI. Neither responded to requests for comment.
“Those who permanently report against Venezuela with hate and envy, they’re jealous of us,” Maduro said, referencing reports about the AI campaign. “… It’s not artificial intelligence. It’s popular intelligence. It’s revolutionary intelligence.”
VTV, the Venezuelan state-owned station, has shown the clips on several programs in recent weeks. After highlighting the videos on Saturday, “La Hojilla” host Mario Silva insulted a journalist who reported on the AI campaign for El Pais, accusing her of wanting to “drag Venezuela through the mud.” VTV and Silva did not respond to requests for comment.
A woman’s TikTok video mocked Venezuelan politicians. She was arrested.
From Twitter ‘army’ to ‘hashtag of the day’
Because Maduro has tightened his grip on media outlets, nonprofits and human rights organizations, many Venezuelans now turn to social media for information.
“That’s in essence what has made the government run to these networks — an attempt at total communications hegemony. They see this as war, and Twitter is their battlefield,” González said.
In 2017, the Ministry of Justice greenlighted a project for “armies of trolls of the Venezuelan revolution,” or thousands of “creators of false information to confuse the opposition,” an investigation by IPYS Venezuela, an organization that promotes free speech, found.
Venezuela tapped 1.5 million phone lines. It’s just the start, experts warn.
Since then, the state has censored a slew of websites and scrubbed online content, investigations found. It has also organized users to “disseminate Twitter messages that espouse support of the government, divert attention from events that could harm the government or benefit the opposition, and fracture opposition groups,” according to a report by the watchdog group Freedom House.
Those users are paid to boost the “hashtag of the day” from the Ministry of Communications and Information, which treats posting like a competition.
One of the most prominent social media campaigns targeted investigative journalist Roberto Deniz because of his extensive reporting on Alex Saab, a government financier who was extradited to the United States on corruption charges. Freedom House found tens of thousands of tweets with hashtags like #DenizExtorsionador — or Deniz extortionist. State security forces raided Deniz’s parents’ home in 2021.
“It’s a massive propaganda machine with a great capacity to bombard and saturate social media,” Deniz, from Armando.info, said. “They use these tools to confuse people inside Venezuela, but also abroad.”
AI porn is easy to make now. For women, that’s a nightmare.
Last year, Nisos — a U.S.-based intelligence firm — uncovered a Venezuelan campaign to drive the social media narrative on Colombia’s presidential elections. Those accounts bolstered then-candidate Gustavo Petro, a Maduro ally, and shared Russian disinformation.
“A lot of the disinformation campaigns that we’ve seen in South American countries have had a Venezuelan component to them,” said Vincas Ciziunas, a Nisos analyst.
The recent disinformation campaigns in Venezuela and China underscore how easy it is to abuse AI, experts said.
“Going forward, it’s going to be so hard to tell what’s real from what’s fake. And we can’t deal with the very fundamental things to be a democratic society when everybody has their own set of facts,” Farid said.
Why are we so afraid of AI?
Potential solutions, Farid said, include platforms labeling content as deepfakes or AI companies watermarking avatars — something Synthesia said it is now doing.
“Ultimately, these companies have got to do better. They can’t keep building this technology, deploying it, privatizing all the profits and socializing the costs,” Farid said.