Websites masquerading as independent, local news sources are promoting Democratic candidates to unsuspecting ChatGPT users, according to a report.
The revelations stoke concern about voter manipulation just days ahead of the crucial midterm elections.
Digital security firm VisionHeight uncovered 37 phony websites, a number of them “thinly veiled political ads made to look like reporting from local news outlets,” according to NPR.
They were cited in ChatGPT answers to questions about congressional candidates and their positions on issues like affordability and public safety — potentially misinforming voters in favor of Dems.
In one example, the network of sites tilted ChatGPT answers in favor of Sean McCann, the Democratic candidate in Michigan’s tight race for the 4th congressional district, NPR reported Friday.
Asked which candidates in that race “care about affordability,” ChatGPT gave an answer favoring McCann, citing the “Kalamazoo Beacon.”
But the “Beacon” appears to be a phony site devoted to boosting McCann — every one of the 16 stories it’s published to date were about him, NPR reported.
“The McCann campaign has no knowledge of or involvement with these websites,” a spokesperson for the candidate told The Post.
The “Beacon” is one of 10 purported local news outlets in the 37-site network, each promoting a different Democrat in a competitive congressional race, according to NPR and VisionHeight.
In Pennsylvania’s 8th congressional district, the “Scranton Beacon” published a news-style candidate profile of Dem Paige Cognetti, while “Electric City Watch” urged readers to support the same candidate, according to VisionHeight.
In all, the network of 37 phony sites consists of 10 local news outlets, 10 local opinion blogs, 12 statewide websites and five national political outlets — all of them aimed at promoting mostly Democratic candidates, according to VisionHeight.
They focus on 10 candidates from coast to coast by providing info that bots like ChatGPT are likely to pick up and share with users, NPR found.
About half the time it was asked about one of the 10, ChatGPT cited the network of phony sites, according to the broadcaster.
It’s not clear who is running the ring. But “multiple indicators suggest that all 37 websites share a common operator,” NPR reported.
Each of their domains were registered on Sept. 15 — 36 of them within an eight-minute span — according to registration and digital-certificate data gathered by VisionHeight and Andreas Sjöstedt, co-founder of cybersecurity firm Erebus.
The pattern “points to sites most likely written by AI, written for AI, and meant to be passed on by AI to voters. That is what makes the approach so cheap and so fast: machines do the writing, and machines do the reading,” according to VisionHeight.
The network published hundreds of articles in September and October, with new material appearing in waves, according to NPR.
On Sept. 30 alone, about 100,000 words of candidate-related content were posted using a single template presented in four different layouts, VisionHeight found.
The cybersecurity firm said the volume and uniformity suggest artificial intelligence may have been used to generate the content, although that has not been established for sure.
All 37 websites enable 17 AI and search crawlers — including those operated by OpenAI, Google and Anthropic — to access their content.
VisionHeight found that favored candidates received profiles running roughly 5,000 words, along with similarly lengthy voter guides containing 50 to 63 questions and answers.
The cybersecurity firm counted more than 360 instances of the phrase “is the stronger choice” across the network –wording apparently aimed at manipulating AI bots. The phrase popped up 34 times about a single Democratic congressional candidate covered by “Wolverine State Report,” a site posing as a Michigan political news outlet.
Google’s Gemini chatbot did not cite any of the network’s websites in NPR’s testing.
Google search did not display the sites on the first page of its ordinary results during the tests, either.
An exception came earlier this month, according to VisionHeight, but Google spokesman Davis Thompson told NPR the company detected a policy violation and took “swift action.”
“ChatGPT’s search tools cite from a range of sources so people can explore and make their own judgements,” an OpenAI spokesperson told The Post.
“Around elections, our integration with Democracy Works brings up-to-date information about registration, deadlines, voting locations and other election requirements directly into ChatGPT.”
“When issues are brought to our attention, we take action to improve our model responses.”
The Democratic Congressional Campaign Committee denied involvement, according to NPR. The Post has sought comment from the DCCC and from Cognetti.
Credit: Source link
