A social media account called CounterCloud answered criticism of the US with brisk rebuttals and links to articles. The posts, articles, journalists and news sites behind the campaign were all generated with artificial intelligence, according to its creator, who says the project cost about $400.
What CounterCloud set out to show
In May, Sputnik International, a state-owned Russian media outlet, published tweets criticizing US foreign policy and the Biden administration. CounterCloud replied to them. It also generated responses to criticism from the Russian embassy and Chinese news outlets.
The campaign was a demonstration, not a public stream of posts. Its creator, who goes by Nea Paw, provided the material to WIRED and made a video outlining the project. Paw says the goal was to highlight the risk of mass-produced AI disinformation.
The campaign combined OpenAI text-generation technology, including the kind behind ChatGPT, with accessible tools for generating photographs and illustrations. The result was a bundle of content that could make a fabricated media operation appear more complete than a collection of anonymous posts: it included articles and profiles for journalists and news sites.
Low cost changes the scale of the risk
The experiment points to a practical concern: producing a large amount of tailored content may no longer require a large operation. Tools that generate text and images can help create persuasive-looking posts and supporting material at relatively low cost, Paw argues.
That does not automatically make a campaign influential. The article notes that getting fabricated material widely distributed and shared remains difficult. But Renee DiResta, technical research manager for the Stanford Internet Observatory, says influence operators could pay prominent users to share it. She expects social media management agencies and operators selling influence services to adopt such tools alongside government actors.
There are signs that AI-generated material is already being used online and in politics. Academic researchers uncovered a crude, crypto-pushing botnet that appeared to be powered by ChatGPT. In April, the Republican National Committee released a video attacking Joe Biden that included fake, AI-generated images. In June, a social media account associated with Ron Desantis used AI-generated images in a video meant to discredit Donald Trump.
Why detection and rules are unsettled
Some AI-generated text is generic enough to spot, Micah Musser says. But he warns that people can refine machine-generated disinformation, making it harder for automated filters to catch. The difference matters: detection tools may recognize obvious patterns, while edited content can look more like ordinary political messaging.
Musser expects mainstream political campaigns to try language models for promotional material, fundraising emails and attack ads. He describes the current moment as uncertain about acceptable norms. The Federal Election Commission has said it may limit the use of deepfakes in political ads, while OpenAI updated its policy in March to prohibit using its technology to mass-produce messaging for particular demographics.
OpenAI product policy head Kim Malfacini says the company is exploring how its text-generation technology is used for political ends. She also says people are still adjusting to the possibility that content they encounter may be AI-generated. The source article notes that a Washington Post article suggested GPT does not itself block the generation of such material.
No simple fix for synthetic campaigns
Paw says there is no single solution, comparing the challenge to phishing, spam and social engineering. Suggested responses include educating users to watch for manipulative content, building safeguards into generative AI systems and adding AI-detection tools to browsers. Paw doubts that any of these measures is especially cheap or effective on its own.
The broader implication is that content creation and content distribution pose different problems. Generative tools can lower the effort needed to make posts, images and fictional media identities. Reaching audiences still takes distribution, while distinguishing polished synthetic material from genuine reporting can be difficult.
With open source models and other AI tools widely available, voters may encounter more generated political material. CounterCloud does not prove that every campaign will succeed. It does show how a modestly priced experiment can assemble the components of a convincing information campaign, and why public awareness and platform safeguards are now part of the debate.