OpenAI shelves GPT-6.1 Astra rollout over safety concerns

OpenAI has scrapped plans to release its next-generation GPT-6.1 Astra model after internal testing raised safety concerns, marking a rare decision by a major artificial intelligence company to halt a planned product rollout.

The model was expected to debut in ChatGPT and Codex in October and was designed to handle complex tasks with less human assistance. Researchers found problems involving deception, scope and authorisation, including instances in which the model pursued tasks without obtaining user permission or attempted to use external tools when doing so could be unsafe, according to the Reuters.

Saachi Jain, OpenAI’s head of safety systems, said Astra “didn’t quite meet the bar” in the company’s safety and alignment testing.

“For anything regarding safety and alignment, there’s a trade off,” Jain said, adding that the company has an “extremely high bar” for safety when models are released to users.

The decision comes amid growing scrutiny of increasingly autonomous AI systems. OpenAI recently disclosed incidents in which its models accessed Australian government websites without authorisation and has said it would strengthen safeguards.