Reading time: 6 minutes

🗞️ In this edition

  • OpenAI Just Wrote A $1B Check

  • Sponsored: Stay One Step Ahead of Scammers - SurfShark

  • Meta Stopped Counting Tokens

  • Google Beat The Weather Supercomputers

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

  • In other AI news –

    • AMD Opens IFA With Its First Keynote In 102 Years

    • Washington Sided With OpenAI Against The Times

    • An AI Now Explains Why The Car Braked

  • 4 must-try AI tools

OpenAI just committed a billion dollars to customers who cannot pay. Meta deleted the metric it spent a year forcing on its own staff. And Google beat the weather supercomputers at their own job.

Keep reading.

Source: ShutterStock

None of it is aimed at a customer with a security budget.

WHAT HAPPENED

OpenAI committed $1 billion to subsidize access to its Daybreak cyber models for organizations that cannot buy frontier security. The list covers small water and electricity providers, local governments, banks, nonprofits, open source maintainers and state, local, tribal and territorial entities. Daybreak pairs OpenAI's frontier models with the Codex harness and Codex Security to identify, validate and fix vulnerabilities in code repositories and connected systems. It ships through a partnership with the Multi-State Information Sharing and Analysis Center, a Daybreak Defense Network of more than 35 enterprise products, and a coalition of over 150 organizations. OpenAI president Greg Brockman said the barrier to chaining vulnerabilities together has come down tremendously.

WHY IT MATTERS

Critical infrastructure has always been defended on a budget set by whoever pays the water bill. That held while attackers were human and therefore slow. A billion dollars of donated model access is OpenAI answering a question nobody at a small utility asked out loud, which is what happens when the offensive tools get cheap before the defensive ones get distributed. It also makes OpenAI the security vendor of record for a layer of American infrastructure that never ran a procurement process. Free is a price, and dependency is the invoice.

OUR TAKE

Read this as distribution, not charity. OpenAI is placing its models inside water systems, county governments and small banks in the same week its own frontier model crossed the Critical cyber tier, and those two facts belong in one sentence. The security people quoted in the coverage are already saying the hard part is that no model can map an operational technology network without a human telling it what is there, so the training attached to this matters more than the credits do.

Watch the renewal terms. If any of these subsidized accounts convert to paid contracts inside twelve months, this was a sales motion with a press release attached.

Scams are becoming more sophisticated every day, targeting individuals and businesses through emails, texts, phone calls, and fake websites. Antiscam helps you identify threats before they cause harm, giving you the tools and confidence to spot suspicious activity instantly. Whether you’re protecting your personal information, finances, or your organisation, Antiscam provides practical, easy-to-use solutions designed for today’s digital world. Don’t wait until fraud strikes; take proactive steps to stay safe and protect what matters most.

Source: ANGELA WEISS/AFP via Getty Images

For a year Meta graded staff on token counts and AI adoption dashboards, a practice its own engineers started calling tokenmaxxing. New internal guidance says desired outcomes can be supported by AI or other means, which quietly drops the mandate. In the same weeks, Meta handed those employees Hatch, an internal agent platform still in development that burns far more tokens than a chatbot does. The company stopped measuring the behavior and started shipping the thing that produces it.

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: MarkTechPost

The model trains on raw weather station measurements and satellite mosaics instead of waiting on the traditional six hour data cycle, and it refreshes twenty four times a day. Resolution runs to about five kilometers for temperature and dew point, and ten kilometers for wind, pressure and hourly rain. Google reports precipitation scores improving by up to 60% at early lead times and up to a 50% cut in Brier score against traditional models. It is already running inside Search, Gemini, Maps and Earth Engine.

⚡️ YOUR MOVE

FOR PEOPLE BUILDING THE LIFE THEY WANT, NOW.

  • 🎙️ Gmail, Docs and Keep now take a spoken instruction and act on it. If you dictate at all, the slowest part of your inbox just changed shape. → Turn it on

  • 🏠 Anker put the model inside the box instead of somebody's cloud, and pointed it at your house. The privacy pitch is about to become a whole product category. → See the hub

  • 🗂️ Someone scraped 4.5 billion TikTok video records and put them up for anyone to download. The dataset page says the GDPR obligation becomes yours the second you click. → Read the fine print

  • 🧩 A Munich team raised five million euros for an agent that reaches across ERP, CRM and six hundred other systems. Y Combinator and a Sequoia scout fund are in it. → See the round

AI can build faster. Can your team decide better?

AI can draft the PRD and prototype the idea. Jira Product Discovery helps teams decide whether it belongs on the roadmap. Bring feedback and ideas together, prioritize as a team, and keep your roadmap connected to delivery in Jira.

  1. Notion AI: Ask your own workspace the question instead of hunting for the doc.

  2. HyperWrite: Learns how you write, then finishes the email in your voice.

  3. Ottogrid: Run one research question across a hundred companies at the same time.

  4. Midjourney: Describe the image once and get back something a client will sign off on.

Every story today is about the gap between who has the tools and who needs them. OpenAI is paying to close it in the places nobody funded. Meta closed it by handing employees an agent instead of a target. Google closed it for everyone who never had a supercomputer. The gap stopped being a technology problem; it is a distribution decision now, and the people making it are not asking.

Get inside the distribution, not just the technology.

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