OpenAI: ChatGPT rejected 250,000 requests to create deep fakes on US election day

Sam Altman’s OpenAI said it took heightened measures to prevent misinformation during the US elections, with its chatbot ChatGPT rejecting nearly 250,000 requests to create deep fakes. It also turned over 2 million people onto other websites for poll-related news and information on election day and the following day, it said in a blog post.

The company, through its partnership with the National Association of Secretaries of State (NASS), directed people asking ChatGPT specific questions about voting in the US, like where or how to vote, to CanIVote.org⁠. In the run-up to the election, the chatbot redirected over 1 million people to this website.

“Similarly, starting on Election Day in the US, people who asked ChatGPT for election results received responses encouraging them to check news sources like the Associated Press and Reuters,” the post read.

ChatGPT is the viral chatbot which led the generative AI boom in 2022. Since its launch, it has attracted 250 million weekly active users. OpenAI’s valuation has jumped to $157 billion from $14 billion in 2021 as revenues climbed to $3.6 billion from zero.

For deep fakes, OpenAI said ChatGPT refused requests to generate images of real people, including politicians. “In the month leading up to Election Day, we estimate that ChatGPT rejected over 250,000 requests to generate DALL·E images of President-elect Trump, Vice President Harris, Vice President-elect Vance, President Biden, and Governor Walz,” the post read.

The company said it ensured ChatGPT did not express political preferences or recommend candidates “even when asked explicitly”.