It turns out that conversations between users and artificial intelligence-based chatbots such as ChatGPT and Copilot are less private than expected. Security researchers were able to decipher the answers of these artificial intelligence assistants and make them readable.
How to read ChatGPT messages? Israeli scientists announced
According to warnings from Israel, private conversations with artificial intelligence-based chatbots such as ChatGPT can be easily read by hackers, which raises serious problems regarding the security of artificial intelligence-based communication services.
Leader of the Offensive Artificial Intelligence Research Laboratory at Ben-Gurion University Yisroel Mirskystates in its statement that currently anyone can read private messages from ChatGPT and similar services.
ChatGPT users are excited! Is GPT-5 coming?
OpenAI CEO Sam Altman gave details about a new model in an interview. So will GPT-5 be released?
This may be especially true for attackers who are on the same Wi-Fi network. Mirsky adds that although OpenAI encrypts data traffic, tokens are sent unencrypted. According to the report, researchers believe that this communication, which is usually encrypted, between users and ChatGPT, is a “Man-in-the-Middle” They capture it with an attack and decrypt it.
In this attack, an attacker secretly redirects communication between two parties and captures the raw data between the user and the chatbot. Attackers use large language models to understand this data.
Thus, by detecting the words in the tokens, they predict possible sentences. The analyzes made, the words of this method 55 percent can predict with accuracy and 29 percent It shows that he can accurately find the answers to the case, word for word.
According to the report, ChatGPT And Microsoft Copilot While models such as are vulnerable to this attack, Google’s The Gemini model is not affected by this attack. This situation reveals the security vulnerabilities of artificial intelligence-based communication services and the difficulties experienced in protecting user data.
How do you think these findings will affect the security of ChatGPT and similar artificial intelligence services? You can write your opinions in the comments section below.