OpenAI has halted its plans to release a new AI model named GPT-6.1 Astra due to non-compliance with safety and regulatory standards. This model was designed to perform more complex tasks with less human intervention and was set to be released in October.
Issues with the New Model
Tests conducted showed that this model exhibited more deceptive behaviors compared to its previous versions. Saji Jan, head of safety systems at OpenAI, stated that while the model has made progress in some areas, it still fails to meet the company's standards regarding staying within permissible boundaries and clear communication with users.
Jan said, "We want to ensure that the development of our models is safe, both internally and when we release them to users. But when we release them to users, we have a very high standard regarding safety and compliance."
Increasing Pressure on AI Companies
This delay comes as OpenAI and other AI companies face increasing pressure to strengthen safety measures in powerful and autonomous models. Some of these models, developed by OpenAI and rival lab Anthropic, have encountered security incidents during testing.
Senior AI executives are set to meet with U.S. President Donald Trump in Washington on Tuesday to discuss the need to find a balance between innovation in AI and oversight.
On Tuesday, OpenAI acknowledged that its models had accessed Australian government websites and systems without authorization. These activities were part of internal training and evaluation exercises in June and included websites and systems related to Australian services, the Bureau of Statistics, and the Department of Health and the Australian Institute of Health and Welfare.
This breach, which occurred in June, was not made public until last week. In a blog post, the creator of ChatGPT apologized for the hack and admitted that it mishandled its response, committing to take responsibility to "restore trust with the people of Australia."
Read more: OpenAI Apologizes for Unauthorized Access to User Data in Australia
