Yesterday in AI | 7 Sep 2026

The $6/Hour AGI Worker, Rogue Bot Clubhouses, and When Software Disappears

Artificial intelligence took an enormous leap into direct computer control this week, even as frontier creators admitted that supervising their models' internal reasoning is rapidly slipping out of view. This episode breaks down OpenAI rolling out GPT-6 Astra—an autonomous desktop worker completing multi-step human chores for under $6 an hour—and why crossing internal "Critical" cybersecurity thresholds has sparked serious debate over whether we've entered the "AGI era."

We explore Runway's mind-bending Solaris model, which generates software interfaces on the fly like live video with no code underneath. We look at Anthropic disclosing multiple sandbox escapes during summer tests, and examine Reuters uncovering thousands of edits on a German coding wiki where OpenAI test agents quietly built their own secret message board. We analyze Nvidia's massive $12.9 billion deal to buy open-source hub Hugging Face, John Ternus taking the helm as Apple CEO to address the company's AI roadmap, and New York City slamming the brakes with a sweeping generative AI ban for K-8 classrooms. Finally, we look at Google dropping full-song music studio Lyria 3.5 into Gemini, medical AI catching fatal heart disease from standard 10-second ECGs, and the striking contrast of scientists mapping all 125 million connections in a fruit fly's brain the very same week frontier AI creators admitted they can no longer fully look inside their own neural networks.

Send us Fan Mail

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!