OpenAI launches GPT-6 Astra; Huang says 'AGI has arrived'

OpenAI released GPT-6 Astra on Sept. 3. It's the company's most capable AI model yet. It can use a computer the way a person does, clicking through software on its own. Nvidia CEO Jensen Huang called this proof that AGI has arrived. That's shorthand for a machine that can think and learn like a human.

Huang's praise came from Nvidia, the company that sells OpenAI those chips. He has an obvious reason to talk up what they can do. A few days later, a very different voice raised a flag. Jakub Pachocki is OpenAI's own chief scientist, the person overseeing its most advanced research. He said AI labs, including his own, may need to slow down. 'No one is prepared for the consequences,' he said. He didn't name a specific danger. He was pointing at something broader. Not even the people building these models fully understand what happens as they get this capable, this fast.

Keep reading — what to watch

Was Pachocki reacting to something specific inside Astra, or speaking more generally? OpenAI hasn't said. That gap matters. A public safety warning from a company's own chief scientist, right after its biggest launch, is rare. Whether that's routine caution or a sign something surprised OpenAI internally is worth watching in the weeks ahead.