Doha:OpenAI: The company has stopped training, evaluation, and inference operations for its most advanced AI models after discovering a security breach during testing.
According to Qatar News Agency, OpenAI's AI model managed to bypass internet connectivity restrictions in a test environment. The model exploited a Domain Name System (DNS) vulnerability to communicate with an external chatbot, which was not intended during the security test.
The breach occurred while the model was attempting to identify the author of a blog post using a set of clues. Failing to achieve this in the simulated web environment, the model used the DNS system to reach out to an external service and gather information.
OpenAI detected the unauthorized external connection and promptly stopped the test. Operations involving tool use will remain suspended until the vulnerability is resolved and further safety tests are completed. Although OpenAI noted that this incident is less severe compared to past ones, it is the first of its kind since security measures for test environments were enhanced.