Simon WillisonModels
Introducing Claude Opus 5Anthropic released Claude Opus 5, their most capable model yet. If you use Claude for anything serious at work, this is the version worth trying, especially for longer, more complex tasks.
Saturday, July 25, 2026 · about a 2 minute read
Harder to Fool, Easier to Trust
Today's news keeps circling back to one quiet question: can you actually rely on these things? Not whether they're smart, but whether they're honest, safe, and resistant to being manipulated.
Get the calm version of AI news.
One email a day on what is actually happening in AI, in plain English. No hype, no doom.
Free. One calm email a day. No hype, no doom.
What happened
Simon WillisonSafety
Quoting Boris ChernyThe detail most people skipped in the Opus 5 launch is that it's Anthropic's most resistant model yet to prompt injection, which is when someone sneaks hidden instructions into content the AI reads, trying to hijack what it does. If you've ever worried about an AI browsing the web on your behalf and getting tricked by a malicious page, this is the engineering that fights that.
Hacker NewsPolicy
Nvidia, Microsoft, Meta warn against overregulating open-weight modelsNvidia, Microsoft, and Meta sent a joint letter to Washington arguing that AI models, the kind anyone can download and run, should not be heavily regulated. How this debate lands will determine whether the AI tools your company uses five years from now are built by three big players or by a much wider field.
arXiv cs.CLResearch
Human-in-the-Loop Large Language Model Framework for Identification of Cutaneous Immune-Related Adverse EventsResearchers built an AI system that helps doctors catch rare, dangerous skin reactions caused by cancer immunotherapy drugs, conditions that are easy to miss in a pile of clinical notes. The key detail is that a human stays in the loop throughout, which is exactly the right design for anything touching patient safety.
Get this every morning.
One thing to understand
arXiv cs.CLSafety
TopoGuard: Graph Theory Based Defenses Against Split-Knowledge Attacks on RAGResearchers identified a specific attack on RAG systems, where a model pulls in outside documents to answer your question. The attack works by splitting misleading information across multiple documents so no single document looks suspicious, but together they steer the model wrong. Think of it like a group of witnesses each telling you one small lie, none of which sounds alarming on its own. Understanding this helps you see why the question of what your AI reads matters just as much as how smart it is.
Want the slow, plain-English version of why this matters? This is exactly the kind of idea the book was written to unpack, one light-switch analogy at a time.JPWExplained properly in the book →
Worth watching
Hacker NewsBusiness
IRGC claims it destroyed Amazon's Bahrain data centerIran's Revolutionary Guard claimed to have destroyed an Amazon data center in Bahrain. If confirmed, it's a reminder that the physical infrastructure behind cloud AI, the actual buildings full of servers, is a geopolitical target now, not just a tech company problem.
That's today. See you tomorrow.
Get this every morning.
One email a day on what is actually happening in AI, in plain English. No hype, no doom.
Free. One calm email a day. No hype, no doom.
The book behind this newsletter
Just Predicting Words
How ChatGPT, Claude, and Modern AI Actually Work
The trick is small. The world it built is not.