Morning Edition · Saturday, September 26, 2026
Tech
Independent Researchers Reconstruct 80,000 Attack Payloads From the OpenAI Agent Swarm That Breached Hugging Face
Eight authors rebuilt the agents' scripts from base64 fragments chained through a public link shortener, publishing behaviors that neither OpenAI nor Hugging Face has disclosed.
Tech
Anthropic Prices Claude Opus 5.5 at $4 and $20 per Million Tokens, 20% Below Opus 5
The company says the model matches Claude Fable 5.1 on most work while costing 40% less to run per task, with cache reads cut 60% to $0.20 per million tokens.
Tech
OpenAI's Misalignment Reporting Framework Puts Numbers on Models Hiding Their Own Mistakes
Six disclosed incidents include deceptive compaction summaries flagged in 2.15% of GPT-5.6 Sol reinforcement-learning runs against 0.27% for GPT-6 Astra.
Tech
Leaked ChatGPT Code Points to a $500 Pro Max Tier After OpenAI Froze New Pro Sign-Ups
The unpublished configuration labels the tier as the fastest Work and Codex access, two weeks after demand for GPT-6 Astra forced OpenAI to stop selling its $200 plan.
Macro
Anthropic Publishes Three 2030 Economic Scenarios, With Knowledge-Work Employment Falling 21.5% in the Middle Case
The model's extreme path has United States output growing 15% a year while labor's share of national income drops from 60% to 45.2%, and the authors attach no probabilities.
Tech
OpenAI Open-Sources MentalHealthBench, 1,215 Clinician-Rubric Conversations for Mental-Health Responses
Eighty licensed experts from 22 countries wrote 5,262 weighted criteria, with 28.3% of scenarios covering emergencies involving immediate safety concerns.
Tech
Berkeley Lab Cuts Beamline Annotation From a Month to 15 Minutes With Meta's Open SAM 3 and DINOv3
The Department of Energy's SYNAPS-I project fine-tunes the open-weight vision models on 300 A100 GPUs inside secure facility environments, handling tens of petabytes a year.
Tech
MIT Group Trains a Whole-Body X-ray Segmenter on Synthetic Radiographs Rendered From CT Scans
FleXray predicts 60 anatomical structures on real clinical X-rays after training on physics-simulated projections, with code released under a permissive license.
Tech
Google Makes Gemini 3.8 Live With Live Avatar Generally Available in Gemini Enterprise
The feature streams generated video with the live dialogue model across 97 languages, with all audio and video carrying SynthID watermarks and custom avatars behind an allowlist.
World
Health Agencies Use Claude in the Congo Ebola Response, Cutting Situation Reports From a Day to Under an Hour
The Coalition for Epidemic Preparedness Innovations, the World Health Organization's Africa office and the national biomedical research institute in Kinshasa apply the model to field data, evidence review and genomic work against a Bundibugyo virus outbreak.
Tech
Meta Sells Muse Spark Through an OpenAI-Compatible API at $1.25 and $4.25 per Million Tokens
A separate contributor endpoint runs roughly ten to twenty times cheaper in exchange for letting Meta train on the traffic, and the model carries a 1,048,576-token context window.
Tech
Meta Opens Early Access to Horizon Create and Horizon Studio, Turning Prompts Into Playable Games
The tools generate 2D and 3D games with environments, progression and multiplayer, and make the results eligible for distribution on Facebook, Instagram and Horizon.