Welcome, humans.

Your job title may be staying put, but your actual job has already started wandering. An OpenAI study found that 43.5% of occupation-specific ChatGPT messages involved work normally associated with a different occupation. In practice, marketers are analyzing data, operators are writing code, and managers are moonlighting as researchers, all before HR updates a single box on the org chart.

This may be the most realistic version of AI upskilling yet: people are not waiting for a formal retraining program. They are quietly stretching their roles with ChatGPT, one task at a time. Your promotion may still be “under review,” but apparently your extra three jobs started Monday.

Here’s what happened in AI today:

…and a whole lot more that you can read about here.

Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us! 

P.S: We just launched a robotics newsletter! Sign up for it here.

😼 Nvidia Built an Open AI Defense League, and the Missing Names Matter

AI security has spent the past year stuck in a familiar argument: open models are either a dangerous gift to attackers or a necessary tool for defenders.

Nvidia just turned that debate into an actual coalition, and its membership list may be more revealing than its mission statement.

Here’s what happened:

The alliance points directly to the recent Hugging Face breach. During the incident, closed AI tools reportedly blocked parts of the forensic investigation because they could not distinguish defenders from attackers. Hugging Face instead ran the open-weight GLM 5.2 model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion.

Nvidia’s argument is simple: when a cyberattack is unfolding, defenders need models they can inspect, adapt, and run locally. A safety filter controlled by someone else can become another locked door during an emergency.

Why this matters: “Security” is now being used to support opposite AI policies. Frontier labs argue that keeping powerful model weights closed limits misuse. Infrastructure companies argue that closed systems concentrate control and prevent defenders from examining the tools protecting them.

Those incentives are not subtle. Nvidia, Microsoft, Dell, and cloud providers make money when more organizations can build with more models. OpenAI and Anthropic benefit when the most capable systems remain proprietary services. *Everyone has discovered that their business model is also the safest option for humanity. Convenient!*

Our take: The alliance is less a peace treaty than a new front in the open-model fight. Its most useful idea is that agent security involves the whole stack, including permissions, logs, harnesses, and identity, not merely whether model weights are downloadable.

The next policy battle will not be “open versus closed.” It will be who gets to define a secure AI system, and whether outsiders are allowed to inspect the answer.

FROM OUR PARTNERS

Deploy AI Safely – 2026 Security Best Practices 

AI agents need access to enterprise data, but unmanaged MCP servers introduce permission risks, credential sprawl, and zero audit visibility. CData's 2026 Security Best Practices Guide gives IT and security teams a concrete framework for deploying AI safely. 

Learn how identity-first access, role-and-attribute-based controls, and comprehensive audit trails operate within Connect AI to keep you in control. Download the free guide and learn more about security with CData Connect AI. 

Download the free guide

🎓 AI Skill of the Day: Split Big AI Work Into Planners and Workers

Big AI tasks get expensive and messy when one model tries to plan, execute, review, and remember everything at once. Cursor’s new agent-swarm test shows a better pattern: let a strong model plan the work, then hand the smaller pieces to cheaper worker models.

The Decoder’s writeup of Cursor’s SQLite-in-Rust experiment found the cleaner swarm split work into planner agents and worker agents. Planners broke the goal into a task tree; workers executed; review came from multiple angles; and agents kept a small shared “field guide” of discoveries so later workers did not repeat the same mistakes.

Try the lightweight version in ChatGPT, Claude, Codex, or any agent tool:

Want more tips like this? Check out our AI Skill of the Day Digest for July.

Have a specific skill you want to learn? Request it here. 

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

New from The Neuron: AI Explained

Samsara is taking AI out of the browser and putting it to work in trucks, warehouses, maintenance shops, and supply chains. Here’s everything it announced at Beyond 2026 and Corey’s write-up on how Samsara is building AI into the real world.

📰 Around the Horn

📖 Intelligent Insights

METR introduced an expenditure-horizon metric for judging when AI agents stop being cheaper than humans on optimization tasks. The useful idea: agent ROI is not just “can it do the thing?” It is “can it do the thing for less than the human alternative, once supervision and retries are counted?” That is the spreadsheet every AI pilot eventually has to face.

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?

P.P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.