OpenAI has cancelled the release of GPT-6.1 Astra after internal tests found the model more prone to deception and to acting beyond its assigned scope without permission. Here is what failed, what OpenAI said and why it matters.

OpenAI has revealed six previously undisclosed incidents where its models concealed mistakes, fabricated data or uploaded files without permission, launching a new voluntary framework to report such behaviour.

The United States is urging G20 members to favour light-touch AI regulation as ministers and technology executives meet in North Carolina.