Researchers discover: ChatGPT can be hacked using another artificial intelligence
A team of cybersecurity researchers recently published concerning findings regarding the safety of advanced language models. According to the report, the researchers managed to penetrate the internal systems of ChatGPT, the popular tool from OpenAI, by using another artificial intelligence model called 'Claude'.
The method used by the researchers is called a "chain attack." In this process, they instructed the Claude model to write complex and sophisticated code designed to bypass ChatGPT's defense mechanisms. The goal was to test whether it was possible to cause the system to perform actions it is supposed to block, such as writing offensive content or exposing sensitive information.
The experiment succeeded in bypassing existing restrictions, which is raising many questions among technology experts around the world. Today, many companies use artificial intelligence for business purposes, including in Israel, and therefore information security in these systems is becoming a critical issue for organizations and private users alike.
Following the publication, OpenAI announced that it is reviewing the findings and working to update its defense mechanisms to prevent similar attempts in the future. Cyber experts emphasize that this is an important reminder that artificial intelligence technology is still in the development stages, and that its use must continue to be monitored to ensure a safe digital environment for everyone.