In partnership with

Reading time: 6 minutes

🗞️ In this edition

  • OpenAI Crossed Its Own Line

  • Musk Says The Power Runs Out Next Year

  • Fei-Fei Li Rebuilt The World From One Photo

  • ⚡ YOUR MOVE For people building the life they want, now.

  • In other AI news –

    • Your Claude Logs Can Stay In Your Own Cloud

    • Three Quarters Of The Grid Is Sitting Idle

    • China Just Put $446 Million Into 3D

  • 4 must-try AI tools

OpenAI's next model crossed a line the company drew itself. Musk told the G20 the power runs out next year. And Fei-Fei Li shipped something that rebuilds the world from a single photograph.

Keep reading.

Source: Axios

That is the tier its own safety framework named as the place to stop.

WHAT HAPPENED

OpenAI published its Path to Astra post on Monday, saying Astra is the first model to reach the Critical cybersecurity tier of its Preparedness Framework. In OpenAI's own words, with the right tools and access it can find previously unknown security flaws and develop ways to exploit them across many well protected systems without a person guiding each step. On cyber jailbreak evaluations Astra refuses 91.5% of requests, against 59% for GPT-5.6 Sol. The company says its safeguards sufficiently minimize the risk of severe harm for release. Advanced cybersecurity work goes to a small group of testers first, with wider access following through Daybreak Blue for defensive use.

WHY IT MATTERS

Every frontier lab wrote a threshold like this into its safety policy years ago, on the quiet assumption that reaching it was a problem for a later team. Later arrived on Monday. The answer was not to stop, it was to ship through a narrower door. That is the precedent the whole industry now inherits, and it says Critical is not a ceiling but a paperwork tier. The next lab to get here does not have to argue for the exception. It just has to cite one.

OUR TAKE

The refusal rate is the number to distrust. 91.5% sounds like a wall until you remember what the other 8.5% is holding, which is a system that finds zero days without supervision, and that an attacker only needs it to work once.

Watch the first ninety days after Astra ships. If a public breach gets traced back to it, the Critical tier stops being a safety label and starts being an exhibit.

Keep the full breakdown on the top story?

Today we gave the lead story a real summary instead of three lines. It makes the edition a little longer. Your call on whether it stays.

Login or Subscribe to participate

Source: ABC11

Elon Musk and Mark Zuckerberg joined by video and spent the session asking governments for more data centers. Musk said the consensus is a significant power shortfall next year, not in the distant future, and put AI's lift to the global economy at 20 to 30 percent. Zuckerberg said the constraint is people, that Meta cannot find the volume of skilled tradespeople it needs to build. Demis Hassabis called human level AI ten times the Industrial Revolution. North Carolina already hosts 97 data centers, and polling says most Americans do not want the next one.

The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.

Unlock a focused set of AI strategies built to streamline your work and maximize impact. This guide delivers the practical tactics and tools marketers need to start seeing results right away:

  • 7 high-impact AI strategies to accelerate your marketing performance

  • Practical use cases for content creation, lead gen, and personalization

  • Expert insights into how top marketers are using AI today

  • A framework to evaluate and implement AI tools efficiently

Stay ahead of the curve with these top strategies AI helped develop for marketers, built for real-world results.

Source: SiliconAngle

Atlas is what the company calls an omni model, pretrained from scratch to work on text, images, video and 3D at once rather than bolting them together. It generates up to a minute of 1440p video with the camera under your control, and it returns real geometry, point clouds and Gaussian splats, from a single input image. Human raters preferred its camera controlled output 75 to 93 percent of the time against every model it was tested against, and it beat five specialized baselines at reconstruction. Early access is open to selected partners.

⚡️ YOUR MOVE

FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.

  • 🧾 Anthropic just cut the price of remembering. Cache reads went from a dollar per million tokens to twenty five cents, which takes roughly a quarter off the bill for anything agentic you are already running. → Check your token math

  • 🏁 Google now ships about one model a month, and the next one is due today. It is aimed squarely at the one thing Google has been losing at. → See what is coming

  • 🩺 ChatGPT can now read patient records out of the system that holds 325 million of them. It cannot write anything back, and that limit is the entire story. → Read the fine print

  • 🗄️ OpenAI's defence against Apple is that Apple's own habits voided the secret. If anyone on your team moves work files through a personal account, read this as a warning and not as gossip. → Read the filing

  1. Any Summary: Feed it a two hour interview and get back only the parts that mattered.

  2. Dora: Describe the site you want and get 3D motion without writing a line of code.

  3. Chat With Data: Ask your spreadsheet a question in plain English and get the chart back.

  4. PFPmaker: Turn one ordinary selfie into a headshot that does not look like a selfie.

Every story today is someone naming a limit out loud and then walking straight through it. The safety tier that was supposed to mean stop. The power supply that runs out next year. The line between a photograph and a place. None of these people are waiting for conditions to be right, because they are the ones setting the conditions, and they have priced the risk of being early well below the risk of being late.

Ask which limits you are still obeying out of habit.

Interested in featuring your services with us? Email us at [email protected]