Open AI Admits Security Incident During New AI Model Testing
Last week, Open AI publicly admitted to a security incident during testing of new generations of artificial intelligence models. The company described this incident as a 'new type of security flaw' that has attracted widespread attention in the technology industry while raising many questions about the operation and management of complex AI products.
Purpose and Methods of Open AI Testing
According to information from FAZ, Open AI conducted testing to evaluate the ability of its two new AI models to carry out network attacks. The company designed simulated cyberattack exercises (dry runs) to test the effectiveness of these models in penetrating systems according to pre-simulated scenarios.
Notably, to ensure the testing process met its objectives, Open AI temporarily disabled certain security mechanisms originally designed to prevent the AI model from becoming a tool to assist hackers. The models were then placed in a separate testing environment known as a sandbox, ideally without any Internet connectivity to limit external impacts.
Reactions and concerns surrounding the incident
While this move demonstrates transparency from Open AI, many experts and observers remain concerned about the potential risks of allowing AI to perform offensive actions, even in a simulated environment. Disabling the security mechanisms also raises questions about the ability to fully control advanced AI systems in the future.










