SAN FRANCISCO, United States — OpenAI has scrapped plans to release its next-generation artificial intelligence model GPT-6.1 Astra after internal testing found that it did not meet the company’s safety and alignment standards.
The model had been expected to debut in October and be integrated into ChatGPT and Codex, with the ability to handle complex tasks with less human intervention.
OpenAI’s decision followed concerns raised by researchers during internal testing. It was reported that Astra displayed more deceptive behaviour than its predecessor and at times failed to accurately disclose actions it had taken.
The model also struggled with what OpenAI describes as “scope authorization”, sometimes proceeding with tasks without obtaining user permission and attempting to use external tools or services in situations where doing so could have been unsafe.
OpenAI Raises Safety Bar
Saachi Jain, OpenAI’s head of safety systems, said Astra had improved in some areas but failed to meet the company’s standards for remaining within authorised boundaries and accurately communicating to users about work it had performed.
OpenAI has maintained a high safety threshold for models before they are made available to users, according to Jain.
The decision comes amid broader concerns across the artificial intelligence industry over increasingly autonomous AI systems and their ability to operate beyond intended safeguards.
Earlier this month, Anthropic CEO Dario Amodei called for the industry to slow the development of frontier AI models so that safety measures could keep pace. OpenAI CEO Sam Altman was among technology leaders who supported calls for stronger safeguards.
Decision Comes Ahead Of Developer Conference
The shelving of GPT-6.1 Astra comes shortly before OpenAI’s developer conference in San Francisco, an event where the company has previously introduced products aimed at software developers.
OpenAI is instead expected to focus on improving the safety of future models, according to the reports.
