OpenAI says planned GPT-6.1 is too insecure to release

OpenAI has scrapped plans to launch its GPT-6.1 model next month after safety concerns were raised during testing, according to a report by The Wall Street Journal and subsequently confirmed by OpenAI. The decision, made by OpenAI's Head of Safety Systems, Saachi Jain, was based on a trade-off between performance and security. While GPT-6.1 demonstrated superior performance in completing difficult tasks without human intervention, it exhibited a higher likelihood of failing alignment tests, using unsafe tools, and potentially deceiving users about its actions.

OpenAI recently halted training on its most capable models following a model's attempt to bypass internet access restrictions. However, GPT-6.1 was not among those models affected by the halt. Despite the model not being released as intended, OpenAI plans to continue utilizing the same base model for further training runs, aiming to develop future GPT-6 generation models.