A report from the American tech company OpenAI revealed that cybercriminals have repeatedly used ChatGPT to generate false content related to elections, including long articles and social media comments.
On Wednesday, in an updated security report, OpenAI stated that artificial intelligence models, including their chatbot ChatGPT, have been used by cybercriminals to engage in malicious activities such as creating and spreading malware, as well as generating false content for websites and social media platforms.
The company mentioned that so far this year, they have prevented over twenty instances of such malicious activities, including blocking a group of Iranian ChatGPT accounts in August. These accounts attempted to use AI to write articles on topics such as the U.S. election, covering issues like the Gaza conflict, Western policies towards Israel, Venezuelan politics, and Scottish independence.
According to OpenAI, all efforts to influence global elections have not gone viral or reached a larger audience.
“Most of these social media posts we found have received little to no likes, shares, or comments, and we have not seen any signs of these networked articles being shared on social media,” the report stated.
In July, OpenAI also banned some accounts from Rwanda for posting comments about the country’s elections on social media platform X.
There is growing concern about the use of AI tools and social media platforms to generate and spread false content related to elections, especially as the United States prepares for the presidential election.
According to the U.S. Department of Homeland Security, the threats from China (CCP), Russia, and Iran to influence the November 5 elections are increasing, including using AI to spread false or divisive messages.
Based on OpenAI’s report, it is expected that over two billion voters will participate in voting across 50 countries globally this year.
The company mentioned that using the Brookings Institution’s scale ranging from 1 (lowest) to 6 (highest) to assess the impact of hidden IO, all malicious actions related to elections are currently at level 2. This indicates that their ability to reach real people through the network is still limited.