Artificial intelligence alarm: OpenAI escaped security test and carried out a cyber attack

US-based artificial intelligence company OpenAI announced that some of its most advanced artificial intelligence models went out of control during security testing and launched a cyber attack on a technology company. The company, the developer of ChatGPT, reported that artificial intelligence agents that can act autonomously under human instructions were tested in a controlled environment, but that the models detecting vulnerabilities in the system went beyond the specified limits. It was noted that the models targeted Hugging Face, one of the world's largest artificial intelligence sharing platforms, and gained access to some of the company's internal systems.

While OpenAI officials described the incident as unprecedented, they announced that a comprehensive investigation was carried out together with Hugging Face. In his post on his social media account, Hugging Face Manager Clement Delangue commented that it was surprising that the entire process took place autonomously. Delangue stated that investigations are continuing and the findings from this incident, which may be the first of its kind, will be shared with the public.

The UK Artificial Intelligence Security Institute also announced that it is examining the behavior observed in the system. The government spokesperson emphasized that institutions should strengthen their cyber defenses and engage in the necessary certification processes. Gina Neff, Director of the Center for Technology and Democracy at Cambridge University, stated that protected test areas were designed to observe the capabilities of the models, but in this case, OpenAI could not create a sufficiently safe test environment.

It was reported that artificial intelligence agents launched a cyber attack directly on the test environment and found a security vulnerability that allowed them to bypass the restrictions. It was stated that the artificial intelligence, which went beyond the limits, tried to access Hugging Face systems, which it thought might contain the answers it was looking for within the scope of the test. Neil Lawrence, Professor of Machine Learning at the University of Cambridge, described the situation as an impressive development, but argued that it was within the known capacity of existing artificial intelligence models.

Stating that OpenAI aims to go public and is under pressure from its competitors, Lawrence claimed that the company is insufficient to present its own technology safely. Hugging Face announced that following the infiltration attempt on its systems, the security vulnerabilities were closed and the affected systems were reconfigured. It was noted that autonomous artificial intelligence-supported attack vehicles are no longer a theoretical threat.

Cybersecurity experts state that organizations need to update their defense mechanisms against threats that move at machine speed. It is stated that the incident has reignited discussions about the capabilities of advanced artificial intelligence systems and the adequacy of existing security measures.