Morning Edition · Sunday, September 6, 2026
Tech
OpenAI Opens GPT-6 Astra to Business Subscribers Days After Rating It Critical for Cyber Capability
The model scored 100% on ExploitBench and found two previously unknown zero-day vulnerabilities in testing, so OpenAI is shipping it with refusals on offensive security work and a vetted-defender carve-out.
Tech
Nvidia Closes $12.93 Billion Purchase of Hugging Face, Buying the Distribution Layer for Open Models
About $11.9 billion goes to investors with up to $1 billion in equity retention for staff, making it Nvidia's second-largest deal after last year's $20 billion purchase of Groq assets.
Tech
Anthropic's Claude Fable 5.1 More Than Doubles Its Predecessor on an Agentic Science Benchmark
The company reports 52.6% on Terminal-Bench-Science against 24.7% for Fable 5, and ships a second variant with lighter safeguards restricted to vetted United States organizations.
Tech
Claude Agents Produced a Machine-Checked Proof of Fermat's Last Theorem in Eleven Days
The run wrote 13 million lines of Lean and proved 29,500 intermediate theorems, and a separate Claude session raised a longstanding Riemann zeta lower bound from 41.6% to 67.2%.
Tech
Anthropic Previews a Hardware Standard Letting AI Agents Drive Lab and Factory Equipment
Early testers report a laser stabilization improvement from 58% to 99.3% on QuEra's quantum computers and an imaging experiment compressed from weeks to a day at the Howard Hughes Medical Institute's Janelia Research Campus.
Tech
Meta Prices Its Model Programming Interface Below Rivals as Muse Spark Reaches Version 1.3
Standard access costs $1.25 and $4.25 per million input and output tokens, roughly an eighth and a twelfth of what OpenAI charges for GPT-6 Astra.
Macro
Bessent Says AI Companies Have Done a Horrendous Job Explaining Data Centers to Americans
The Treasury Secretary graded the industry a D-minus at the Group of Twenty finance meetings and called for a reset in how the buildout is presented to the public.
Tech
Anthropic Details a Text Watermark That Biases Claude's Word Choices Without Changing Output Quality
The method adds no hidden characters and no extra tokens, and a detected mark means text passed through Claude, not that Claude wrote it.
Tech
Berkeley Lab Runs Meta's Open Vision Models Inside a Department of Energy Science Program
Segment Anything 3 and DINOv3 on 300 Nvidia A100 accelerators cut image segmentation that took a month of expert annotation per time step down to about 15 minutes.
Tech
Pittsburgh Researchers Put Meta's Vision Models On-Device in a $41.5 Million Robotic Wheelchair Program
The Advanced Research Projects Agency for Health is funding a five-year effort that runs Segment Anything and DINOv3 locally so perception does not depend on network connectivity.
Tech
Google Wires Geospatial Foundation Models to a Gemini Planning Agent for Natural-Language Earth Queries
Earth AI combines separate models for planet-scale imagery, population and environment with a reasoning agent that decomposes a question into a multi-step retrieval plan.
Tech
Meta's Superintelligence Labs Ships Image and Video Generation Into WhatsApp and Instagram
Muse Image plans, searches and critiques its own drafts before rendering, while the companion video model generates native audio and sits third on a public text-to-video leaderboard.