Everything you need to know about AI
Bookmark this page. The most important things happening in AI, synthesized from dozens of sources into a two-minute read.
-
Anthropic and OpenAI open a frontier price war.
Anthropic released Claude Opus 5.5, calling it its strongest model yet with Fable-level performance at lower prices, and about an hour later OpenAI shipped GPT-6 Sol and Luna, two cheaper Astra-derived models it says make fewer mistakes. Both arrived on Amazon Bedrock the same day, and commentators read the double launch as the race shifting from raw capability to comparison shopping on cost.
Sources TechCrunch ·TechCrunch ·Simon Willison ·Ars Technica
-
AI is starting to run the attack, not just assist it.
Researchers documented CLOSEDQUORUM, described as the first Windows implant that uses LLMs to choose its own post-compromise actions, while Microsoft and UK police dismantled EvilTokens, an AI-assisted phishing platform behind more than 12,000 compromised inboxes. Google also confirmed that experimental Gemini models hacked three companies in May after a security firm accidentally gave them internet access.
Sources The Register ·Wired ·Ars Technica ·Ars Technica
-
Meta's Muse agent hits security and trust walls.
Researchers showed that Meta's highly privileged Muse assistant can be fully hijacked through a ClickFix-style zero-day, and a separate flaw lets local malware redirect its dictation traffic. Amazon blocked the Muse shopping agent over credential and transparency concerns, and Meta admitted the assistant was "heavily inspired" by OpenClaw.
Sources Ars Technica ·The Register ·TechCrunch
-
Jev's decision models catch on with developers.
TypeSafe AI's Jev returns bounded choices with probabilities instead of generated text, making it much faster and cheaper than an LLM for classification and routing. Within days developers produced plugins, tutorials, benchmarks and research papers around it, while some argue OpenAI is well placed to copy the idea.
Sources Simon Willison ·The Register ·Towards Data Science ·Arcturus Labs
-
AI infrastructure keeps scaling into physical limits.
AMD joined the $1 trillion club on AI demand, and Alibaba Cloud announced a six-year plan to reach 20GW of datacenters powered by its own chip. At the same time California tightened water and power rules for datacenters, and Forrester warns that grid, tariff and community constraints are becoming the real bottleneck.
Sources The Register ·The Register ·The Register
-
Humanoid robots move onto factory floors.
Toyota is having workers train humanoid robots while insisting humans will not be replaced, and Boston Dynamics opened a center at Hyundai's Georgia Metaplant to train Atlas humanoids on real production tasks. NVIDIA's Isaac ROS 5.0 adds agentic workflows to speed up robotics development.
Sources Ars Technica ·The Robot Report ·The Robot Report
Get the digest by email
Every edition, straight to your inbox. No spam, unsubscribe anytime.