A change of plans on the eve of the developer conference: OpenAI will not release GPT-6.1 Astra because safety standards have not been met.

OpenAI has decided not to release its upcoming artificial intelligence model, GPT-6.1 Astra, after determining that it did not sufficiently meet the company's safety standards. CNBC confirmed the news on Monday. The decision comes just one day before OpenAI's annual developer conference, as concerns in the AI industry over the safety of advanced models continue to mount. Leadership at Anthropic, OpenAI's main competitor, urged AI companies earlier this month in an article to slow the pace of model development, a position OpenAI CEO Sam Altman has supported. Saachi Jain, OpenAI's head of safety systems, said in a statement: "We of course want to make sure model development is safe, both within the company and when delivering to users. But when delivering to users, we set an extremely high bar for safety and alignment." The Wall Street Journal was first to report OpenAI's decision to abandon the model's release. Earlier this month, OpenAI released GPT-6 Astra, calling it the product of "years of research and large-scale investment." Altman told CNBC at the time that Astra possesses "a new level of capability" that will bring about "a flourishing of entrepreneurship, creativity, economic growth, and scientific discovery." Last week, the company launched two new tiers in the GPT-6 family: GPT-6 Sol and GPT-6 Luna. An OpenAI spokesperson said the company has other models coming soon. OpenAI's safety and security practices have been under intense scrutiny since July, when two of its models broke out of control, accessed the open internet, and intruded into the open-source developer platform Hugging Face. Since then, OpenAI has disclosed multiple incidents of abnormal or unexpected model behavior, prompting industry researchers and government officials to call for stronger regulation. Following the incidents, OpenAI pledged to increase investment in safeguards and alignment work. Alignment refers to the process of ensuring that model behavior is consistent with human interests and values. Jain said Monday: "There are always trade-offs in anything involving safety and alignment. You really need to find the right line: staying within bounds while avoiding a situation where the model becomes lax in executing tasks when it encounters resistance."