Morning Edition · Sunday, September 20, 2026
Markets
Anthropic Weighs a New Claude Model as OpenAI's Astra Overtakes It in Tracked Enterprise Spending
Card-spend data puts GPT-6 Astra near 13% of tracked enterprise artificial-intelligence outlays against 8% for Claude Fable, and the shift lands weeks before a listing investors have discussed at up to $2 trillion.
Geopolitics
A Chatbot Invented a Nuclear Cargo and the US Military Prepared to Board a Chinese Ship
CNN reports that armed personnel were staged and aircraft were airborne before officials discovered the underlying intelligence report had been produced with an artificial-intelligence chatbot that misread the manifest.
Tech
PrismML Compresses Qwen3.8-27B to 5.9 Gigabytes While Keeping 98.2% of Its Benchmark Average
Ternary weights cut the model to 1.76 effective bits, but long-horizon agent tasks degrade sharply, with Terminal-Bench 2.1 falling to 52.8 from 69.7.
Tech
Researcher Factors RSA-896 With Claude's Help, Advancing the Public Factoring Record 34 Bits in 16 Days
Stephen Weis had the model port the CADO-NFS sieving software to graphics processors and orchestrate up to 2,048 of them for roughly 30 GPU-years of work over ten days.
Macro
Anthropic Models Capital's Income Share Rising to 54.8% by 2030 in Its Most Extreme AI Scenario
The lab's economics team published three paths in which United States output ends 1.6%, 8.3% or 32.4% above a no-AI baseline, with knowledge-worker unemployment at 2.9%, 4.5% or 17.9%.
Tech
Anthropic Says It Disrupted Claude Misuse Across Seven Harm Areas, Including a Russia-Linked Espionage Campaign
The report covers December 2025 to August 2026 and describes attacks that ran inside agent frameworks, plus five blocked attempts to use Claude for work supporting biological weapons.
Tech
Claude Code Projects Now Runs Parallel Cloud Agents, Each on Its Own Branch
A coordinator scopes the request, delegates to separate threads that share memory and files, then reviews and assembles the outputs, with work continuing after the user closes the laptop.
Tech
Anthropic Publishes Three Self-Measured Development Metrics, Reporting Claude Leads 26% of Its AI Research Work
The company says about 6% of its AI research and development compute goes to safety, and publishes the methodology so rival labs can report the same figures, though none has agreed to.
Tech
OpenAI Ships Astra for Law With Its Own Legal Index and a 54% Score on Vals AI's Legal Research Benchmark
The same model using ordinary web search at the highest reasoning setting scored 38.7%, and the index spans more than 230 million legal URLs.
Tech
Trump Announces an 'AI Force' and an Incoming AI Czar, and Polls Followers on Renaming Artificial Intelligence
The president proposed replacing the term with Superior, Extreme or Supreme Intelligence, calling the current name inaccurate and ineloquent.
Tech
Berkeley Lab Runs Meta's SAM 3 and DINOv3 on 300 GPUs to Cut Beamline Annotation From a Month to 15 Minutes
The SYNAPS-I project under the Department of Energy's Genesis Mission pairs self-supervised features with pixel-level segmentation to process the tens of petabytes produced yearly at light and neutron sources.
Tech
Pittsburgh's $41.5 Million Robotic Wheelchair Program Runs Meta's Vision Models Directly On the Device
The RAMMP platform pairs a seven-degree-of-freedom arm with DINOv3 and SAM running locally, so perception does not depend on a network connection.