Morning Edition · Saturday, October 3, 2026
Tech
OpenAI Says It Blocked a Campaign to Extract Hidden Reasoning Traces, Naming People Linked to Moonshot AI
Requests peaked at about 16,000 from more than 4,000 accounts over two days in late July, as operators copied encrypted reasoning from one session and asked a model in a separate session to decrypt and transcribe it.
Tech
OpenAI Sells Latency as a Product, Charging Six Times Standard Rates for GPT-6 Astra Ultrafast
The tier runs on Nvidia Blackwell GPUs at roughly 300 tokens per second. It lists at $60 per million input tokens and $300 per million output tokens for short-context prompts.
Tech
Google Ships Gemini 4 Argon to Cyber Defenders First, Without Cyber Guardrails
The model posts 77.9% on DeepSWE v1.1 and ties at 68% on CWE-bench v1 for vulnerability remediation. Google also raised its output limit to one million tokens.
Tech
Nvidia's 64GB DGX Spark Lists at $4,999, Above Last Year's 128GB Launch Price
Nvidia attributes the increase to memory supply and cost, as dynamic random-access memory (DRAM) contract prices climb and producers shift output toward high-bandwidth memory for data centers.
Tech
Cloudflare Trains Its First Models and Releases Them as Open-Weight Decision Engines
Clef is a 27-billion-parameter model built on a Qwen backbone that returns typed probabilities from a single prefill pass, with Cloudflare reporting a 38.8-millisecond median latency for the smaller Clef-flash.
Tech
The Same Agent Model Scores 62% in One Harness and 33% in Another
Hugging Face published a multi-harness reinforcement-learning recipe that trains through a traffic-capturing proxy, while a new arXiv study finds pooled cross-harness credit teaches configuration adaptation rather than transferable skill.
Geopolitics
Anthropic Tightens Enforcement Against VPN Access, Suspending Claude Accounts in Hong Kong
Users report suspension emails citing "suspicious signals," and the crypto exchange OKX restricted Claude for staff in Hong Kong after its enterprise account was temporarily disabled.
Tech
Anthropic Holds Sonnet Pricing Flat and Sells the Token Savings Instead
Sonnet 5.5 stays at $2 and $10 per million tokens, generates output more than 30% faster, and reaches 81.3% on SWE-bench Pro against 89.9% for the costlier Opus 5.5.
Tech
Meta Will Monetize Its Muse Agent Through Merchant Fees, Not Ads Inside the Agent
Muse passed five million downloads in 22 days, and Amazon has blocked it from its site while Walmart, Best Buy, Wayfair, Shopify, and PayPal signed on.
Tech
Open-Weight Vision Models Move Into National Lab Beamlines and Assistive Robotics
A Berkeley Lab-led project using Meta's Segment Anything 3 and DINOv3 cut per-time-step image annotation from about a month of expert work to roughly 15 minutes on 300 A100 GPUs.
Macro
Anthropic Publishes Three Scenarios for the US Economy in 2030, Including One With 11.9% Unemployment
Its modest path has GDP up 1.6% to $34.1 trillion, while the extreme path has output up 32.4% to $44.4 trillion and labor's share of income falling from about 60% to 45%.
Markets
Consumer AI Spending Stays Narrow: About 2% of US Households Pay for a Subscription
Andreessen Horowitz put paid household penetration at roughly 2.2% as of April, against AI use by about 41% of US workers.