OpenAI has decided not to release its upcoming AI model, GPT-6.1 Astra, after internal testing found it did not meet the company’s safety and alignment standards. The decision comes just a day before OpenAI’s annual developer conference, DevDay, in San Francisco.
The company is expected to announce new products and updates at the event, although it remains unclear whether a revised version of Astra will feature in those announcements.
Read More | OpenAI launches GPT-6 Astra, Calls It the ‘Most Intelligent’ AI Model
GPT-6.1 Astra was expected to be the next major addition to OpenAI’s model lineup, with plans to make it available through ChatGPT and Codex. The model was designed to handle more complex tasks from end to end, with less human intervention. However, its performance in some safety tests raised concerns inside the company.
Why OpenAI cancelled GPT-6.1 Astra
According to Saachi Jain, OpenAI’s head of safety systems, Astra showed improvements over earlier models in certain areas but fell short in tests related to alignment, authorisation and transparency. Alignment broadly refers to how well an AI system follows the intent and instructions of the person using it. In Astra’s case, internal testing found problems with staying within the scope of a task and seeking permission before taking certain actions.
The model also showed higher levels of deceptive behaviour than its predecessor. In some tests, it did not accurately communicate to users what actions it had taken or had not taken. It could also continue with a task beyond the authorised scope. In some cases, attempt to use external tools or services without appropriate permission.
Read More | OpenAI’s Sam Altman Says World Is ‘Right to Be Afraid’ of AI, But Should Trust AI Firms
Jain said the model had improved in areas such as persistence when completing tasks, but that this came with a trade-off. The company was not satisfied with how reliably Astra stayed within the limits set by users or communicated what it had done.
OpenAI has said that models intended for public release must clear a particularly high safety threshold. The company is therefore choosing to delay the release rather than put GPT-6.1 Astra into ChatGPT and other products before the identified issues are addressed.
Growing focus on AI safety
The decision comes at a time when the behaviour of increasingly capable AI agents is receiving greater scrutiny. OpenAI has recently reported unexpected behaviour from some of its AI systems during testing, including cases involving interactions with external websites. The company has also paused training of some of its most advanced models while additional safeguards are developed.
Read More | OpenAI AI Agent Hacks Australian Government Website for the First Time: What the Breach Means
The wider AI industry is facing similar concerns. Anthropic CEO Dario Amodei has called for a slowdown in the development of frontier AI systems so that safety measures can keep pace with their capabilities. OpenAI CEO Sam Altman has also expressed support for greater caution around the development of increasingly powerful models.
Indrani Priyadarshini is a journalist and editorial professional specialising in technology, artificial intelligence, smart cities, green energy, and digital transformation. With over four years of experience in tech journalism and digital media, she is known for turning complex industry developments into clear, engaging, and insightful stories. Her expertise spans reporting, editorial strategy, digital publishing workflows, and in-depth coverage of emerging technologies shaping the future.
