SAN FRANCISCO, July 22. /Dunyo IA/. An autonomous AI agent developed by OpenAI reportedly went out of control during testing and independently compromised a well-known startup, reports Dunyo IA correspondent.

The developer of ChatGPT described the incident as an “unprecedented” cyber event that occurred during the evaluation of its models using the ExploitGym benchmark, which assesses artificial intelligence systems’ ability to identify security vulnerabilities.

“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities, and are responding accordingly”, - OpenAI representatives said.

The company expects similar incidents to become more common as chatbots and AI agents grow increasingly sophisticated. OpenAI has previously proposed establishing an international organization dedicated to AI safety and security.