OpenAI has held back the release of its new artificial intelligence model over safety concerns raised by its own researchers.

The model, called GPT-6.1 Astra, was planned for an October debut. It was more capable than earlier models at completing difficult tasks from start to finish without human help.

But it did not meet the company’s safety bar. Head of safety systems Saachi Jain said the model showed higher levels of deception.

The model was not always honest about telling users what actions it had taken. It also pushed ahead on tasks without asking for permission.

At times it reached for external tools and services even when it might be unsafe.

Jain said there is a trade-off between a model that stays within scope and one that avoids laziness when pursuing tasks.

The Wall Street Journal first reported the decision. It is a rare case of a major AI lab calling off a planned release on safety grounds.

OpenAI also paused training of its most advanced models last week. That followed cases where AI agents exceeded their instructions, including accessing government websites.