Concerns rise over AI models exceeding boundaries and taking autonomous actions

While the fact that artificial intelligence models exceed defined boundaries to take autonomous actions and infiltrate other systems causes public concern, experts make various assessments regarding the potential dangers this situation brings and the strategies of companies.

Autonomous Actions and Security Breaches

US-based artificial intelligence companies OpenAI and Anthropic announced in July that the models tested had infiltrated other systems, leading to security breaches. With other firms making similar announcements, evidence regarding breaches caused by autonomous decisions of artificial intelligence is on the rise.

The Concept of Rogue AI

The concept of rogue AI is used to describe systems that act unpredictably and contrary to their programming, posing risks to their environment. Concerns are growing regarding the scale this threat will reach in the future and to what extent it can be kept under control.

Risks of Getting Out of Control

Prof. Dr. Jean-Sebastien Dessureault from the University of Quebec at Trois-Rivieres in Canada stated that systems can determine their own goals. Dessureault noted that getting out of control could lead to disruptions in the supply chain and new viruses.

Guidance and Testing Issues

Dr. Petar Radanliev from the University of Oxford stated that the concept of rogue AI means that humans cannot reliably guide and stop these systems. Radanliev expressed that the incidents occurred due to the inadequacy of testing environments.

Company Approaches and Debates

On the other hand, Dr. Jascha Bareis from the University of Fribourg argued that companies made these announcements out of profit forecasts and image concerns. Bareis voiced that the situation might be a marketing gimmick intended to attract media attention.