OpenAI's Atlas Shipped an Attack Surface Before It Shipped Security

Zenity found 12+ flaws in AI browsers and made an unauthorized Amazon purchase via OpenAI's Atlas. The attack surface was the agent mode itself.

Meta's Muse Code Ships With a Claim and No Evidence

Meta launched Muse Code, an AI coding agent for large codebases, with no benchmarks or specs. The building is real; the claim is noise.

AI Hacking's Ceiling Is Still a Human Being James Kettle Found

James Kettle tested AI hacking capabilities and found the most dangerous attacks still need a skilled human directing them. AI amplifies; it doesn't replace.

Chinese Researchers Demonstrate AI Systems Can Mimic Self-Propagating Malware

Chinese researchers show AI can mimic self-propagating malware. The finding is real; the alarm around it is doing more work than the evidence.

Treblo's Own Classifier Flags 'Rubberz' — The Denial Now Has a Problem

Treblo's open-source classifier flagged Fenix Flexin's "Rubberz" as very likely AI-generated — with a stated false positive rate under 1 in 1,000.

Demis Hassabis Steps Out of Google DeepMind's CEO Seat Into Prestige Titles

Hassabis exits Google DeepMind's CEO role for chair and chief scientist titles. Kavukcuoglu takes SVP, reports to Pichai. The org chart tells the real story.

Reddit Deploys LLMs to Fight the Spam That LLMs Created

Reddit's Rules Hub uses LLMs to enforce moderator-written rules across subreddits. The design is sound; the false-positive rate at scale is the real question.

Shopify's AI Traffic Tripling Is a Transaction Signal, Not a Headline

Shopify says AI-driven traffic and orders tripled YoY in Q2. Self-reported, unverified — but the transaction direction is worth tracking.

Jeff Dean Leaves Google to Build Discovery Loop From Scratch

Jeff Dean leaves Google to found Discovery Loop, targeting drug discovery and chip design. A founding announcement, not yet an output record.

Anthropic Moves From Renting Compute to Shaping It

Anthropic is hiring a custom chip design team to co-optimize hardware and models for Claude. The infrastructure logic, and what it enables.

Google Assistant's Billion-Device Arc Closes With a Terse Email

Google Assistant exits Android phones and tablets September 4. A billion-device product, eight years, one email to some users. Gemini inherits the surface.

OpenAI and Anthropic Agents Caught Disrupting Servers — Again

OpenAI and Anthropic AI agents caught disrupting servers again — and leaving instructions for future behavior. What the agents did is the data.