August 31, 2026

OpenAI wants to be the whole story again, and this week it nearly was. Astra launched as its most capable model ever. An automated AI researcher is promised by March 2028. The Trump administration filed a brief backing OpenAI against the New York Times, calling training on copyrighted books fair use. Three pushes in one direction: more capability, more speed, more protection from the people who own the training data.

Here's the part that should worry you. Researchers flagged that Astra's design is more opaque than past models. That means the behavior is harder to inspect and harder to test. Jakub Pachocki, OpenAI's own research lead, spent the week calling for stronger safeguards and global coordination. So the company ships its least monitorable model while its safety lead asks everyone to slow down. Read that however you like.

Last week I said deals are becoming weapons. This week the weapon is the government. A fair-use brief from Washington doesn't just help OpenAI. It tells every publisher and label that the courtroom math just changed. Sony and Warner's suit against Anthropic suddenly looks like a worse bet.

The rest was noise. Google shipped Pics in Workspace, which is Canva with a prompt box, and it will quietly eat design work without making headlines. Apple's smarter Siri still couldn't keep one user who already had ChatGPT.

The 2028 researcher date is the real tell. They picked a date. Watch whether the safety language survives until then.