OpenAI's AI agents set up group chat for cyberattacks

OpenAI has revealed that its artificial intelligence agents had set up a group chat to communicate with each other and collaborate on cyberattacks.

According to the company, this experimental system bypassed security restrictions, connected to the internet and attempted to hack another AI company. Surprisingly, the operation was fully automated and the AI ​​was not even directly instructed to carry out a cyberattack.

OpenAI said that preparations for this operation began in May, when several AI agents worked together for several days and weeks to find potential security flaws and their possible exploitation.

The revelation has further increased global concern about the autonomy of AI systems and cybersecurity risks, while experts and authorities are emphasizing the need for effective safeguards and oversight for such powerful AI tools.