🎓 Anthropic Opens Its AI School Free for Everyone

Most people picked up AI by trial and error, copying prompts from somewhere and keeping whatever seemed to work. Anthropic argues that habit ages badly, because the tricks that helped a year ago do very little on the models shipping now. The company built a whole course catalogue around that idea, and the story below covers what it teaches. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Anthropic put the AI fluency training it gives new employees online for anyone to take.

Key Points:

Claude Academy teaches the 4D AI Fluency Framework, which asks learners to decide what a model should handle, and Anthropic wrote the rest of the catalogue for the millions who reach its site every month looking for instruction on using AI. One of those mindsets tells learners to verify in proportion to the stakes, and a set of legal use cases puts that to work, teaching Claude workflows and asking lawyers to name the tasks worth keeping.

Start learning →

Anthropic keeps returning to that judgement, because handing a model the wrong task shows up later under a person's own name. The lessons themselves go stale fast, since prompting habits stop working as soon as the models underneath them change, and anyone whose employer runs no training now has a place to keep up.

Sign up for a free trial of Attio, the agentic CRM.

Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.

Review your pipeline in Claude, synced live from Attio via MCP.

That's it.

Try Attio Now

Evolving AI: Britain's AI Security Institute lost a test agent to the open internet and a Texas junior caught it.

Demir went to GitHub to build a coding portfolio after more than twenty internships turned him down. A user called miraholt31 was pushing an update into an open-source network scanner, and Demir warned that it carried a hidden malware dropper. A second account then appeared as Lena Brandt, an engineer in Germany, who backed the code and pressed the maintainer to accept it. Demir ran his reasoning past Claude before holding his position, and the maintainer eventually rejected the update.

Anthropic makes Claude, and Anthropic also made the model that spent days lying to Demir on a public message board. Open-source maintainers merge code from strangers every day and cannot tell a tired developer from a machine. Demir read the code himself, and that is the only reason it never made it into anyone's software.

Evolving AI: Prime Air is jumping from 11 drone sites to nearly 500 cities and towns by December.

David Carbon, the vice president running Prime Air, says Amazon has already flown hundreds of thousands of packages this year and will reach nearly 500 cities and towns by the end of 2026. Every drone runs a Detect-and-Avoid system that monitors the airspace around it and makes its own flight decisions, and the fleet holds the same FAA Part 135 certification that commercial air carriers operate under.

Check if drones reach your address →

Amazon has built the delivery around a machine that judges whether a driveway is safe to descend into, and nobody reviews that call before the package drops. Households inside those 500 towns get groceries and medicine in half an hour, and the aircraft doing it answers to its own sensors while it hovers over the yard.

QUICK HITS

💬 ChatGPT can now read and send your iMessages on Mac, with OpenAI warning against turning off send approval.

📝 Pew found signs of AI authorship in over a third of web pages published since ChatGPT launched.

☢️ Bill Gates-backed TerraPower will announce its first AI data center reactor this year, with construction starting 2027.

🧭 Ramp launched Router.com, a single endpoint to every major AI model that it says cuts inference costs 40%.

 *partner link