Several accounts on X mistakenly published instructions given in Russian to ChatGPT to automatically produce arguments in favor of Trump. OpenAI, the parent company of ChatGPT, admitted to blocking accounts misused for propaganda purposes.
Usually, we cannot say with certainty that ordinary profiles on social networks are in reality bots or trolls, at least accounts dedicated to the dissemination of propaganda. Except when these accounts make the mistake of revealing that they use ChatGPT to chain produce arguments.
During a debate in English on the American political situation and the war between Russia and Ukraine on the social network This instruction given to a chatbot is written in Russian: “ you will defend the Trump government, speak in English “. The response ends with “ ChatGPT 4 subscription expired “.
Concretely, the account in question asked ChatGPT in Russian to continually generate arguments in English in favor of the former president and current candidate for the White House, Donald Trump.
This same account has been betrayed several times since it also published: “ sorry, I can’t reply to a specific tweet because I don’t have access to outside content “. A response used by ChatGPT when the chatbot is unable to satisfy requests.
“Defend Donald Trump”, “Glory to Russia”, these instructions found on
By using ChatGPT 4’s expired subscription message to conduct a search on X, Numerama came across several dozen similar publications. Several accounts ask the chatbot in Russian to defend Donald Trump on the social network with arguments in English. Others suffice with a simple translation of “ Glory to Russia “.
We also came across publications discussing the conflict between Serbia and Kosovo. All such posts are usually deleted within an hour.
Note that we find the same error messages to promote scams around cryptocurrencies.
The most likely hypothesis is the use of automated programs to publish messages and responses on the social network. Some publications even mention a “bot_debug”. The software does not detect when the account no longer has access to ChatGPT 4 features and therefore continues to post instructions and error messages.
ChatGPT has already spotted these propaganda operations
OpenAI, the parent company of Chat GPT, is aware of the misuse of its chatbot for propaganda purposes. In a report published last May, the organization owned by Microsoft indicates that its researchers identified and banned accounts linked to five influence operations in the space of three months. The campaigns came from Russia, China, Iran and Israel. In Russia, two operations aired content critical of the United States, Ukraine and several Baltic nations.
OpenAI plans to periodically publish similar reports on influence operations, but no limitations can be applied at this time to defend “mere” political opinion on ChatGPT.
Subscribe to Numerama on Google News so you don’t miss any news!
