[01]
OpenAI Says New Model Meets Its 'Critical' Cybersecurity Threshold
OpenAI says its upcoming Astra model is the first to cross its "Critical" cybersecurity capability threshold.
// daily-digest · wed 2 sep 2026 · morning refresh · last 72h focus
Nine items on frontier-model safety thresholds and new releases, Stripe's widening footprint, and a Rails agentic-coding benchmark, plus a wildcard on coffee and bone density.
— refreshed for you, in 9 items.
OpenAI says its upcoming Astra model is the first to cross its "Critical" cybersecurity capability threshold.
Anthropic released Claude Fable 5.1 and Mythos 5.1, cutting cache costs and improving coding and research benchmarks.
Perplexity launched a hybrid compute feature for Mac that splits AI tasks between cloud and local models.
Fortune compares Stripe's acquisition-driven expansion into AI and blockchain to Google's early-2000s buying spree.
Google is replacing Gemini Notebook's fixed daily usage caps with limits based on prompt and feature complexity.
A Rails agentic-coding benchmark found Claude Fable 5.1 matches Opus 5's accuracy at roughly half the cost.
Paint.NET's creator describes using Claude to build a from-scratch Direct2D reimplementation enabling Linux compatibility via WINE.
Sean Goedecke outlines tactics for handling AI-generated "workslop" colleagues send, given the asymmetric effort to review it.
A decade-long study of nearly 10,000 older women found heavy coffee drinking linked to lower hip bone density.