Today: Anthropic and OpenAI tighten security controls while automatic coding tools reach solo founders. Alibaba stretches AI video to 30 seconds, and physical AI devices start to take shape.
Anthropic turns on auto mode in Claude Code for Pro, Max and Team plans starting August 14. Instead of bugging you with confirmations, the system spots and blocks destructive commands on its own. In Anthropic's internal tests, teams shipped 25% more code.
Read onOpenAI's next model hit a cybersecurity warning level: it could find and build exploits in real-world systems without a human in the loop. The company is setting up isolated test environments and working with governments to check these abilities before launch.
Read onMatt Wolfe posts his video breakdown of the week's key AI news.
Read onClaude's safety filters now do a better job telling when a biology question is legit and when it's risky. Needless blocks dropped 85%, while protections stay in place for high-risk topics like housing or toxicology.
Read onWan3.0 is Alibaba's new video generator. It turns text, images, PDFs and even slide decks into smooth videos up to 30 seconds long. It takes many kinds of input and keeps the visuals consistent. If you already have slide decks or PDFs about your services, try turning them into short videos.
Read onComputer for Builders runs 15+ AI models to write, deploy and monitor code, track payments and catch fraud. It connects to GitHub, Stripe and more. It's available to Pro and Max subscribers.
Read onA smart speaker the size of a hockey puck, with no screen, that would run ChatGPT and arrive in 2027 for more than $300. It has a camera, sensors and moving parts that react when it talks.
Read onAI systems generated new synthetic viruses nature has never seen. That raises questions about biosecurity and how to regulate these tools.
Read on