Morning Edition · Saturday, October 10, 2026
Tech
OpenAI's unreleased model claims 372 math result families, and specialists say most remain unverified
Lean formalizations cover the main result of only 162 of the 722 manuscripts, and a group of mathematicians, in a statement Fields medalist Terence Tao posted to his blog, has urged colleagues to stop working with OpenAI.
Tech
Fired OpenAI safety researchers deny leaking reports on Astra's less monitorable reasoning
OpenAI says Jasmine Wang, Tomek Korbak and Mikita Balesni showed a "pattern of misconduct" in handling research information, and the three say their work on monitorability required talking to outside groups.
Tech
Anthropic prices Claude Haiku 5.5 at one-tenth of Haiku 4.5 and lifts its OSWorld score to 72.4%
Input costs $0.10 per million tokens for prompts up to 100,000 tokens, every rate is five times higher above that, and the model accepts a one-million-token context with adjustable effort.
Tech
Microsoft's Decision-1 scores fixed answer options in one pass and charges nothing for output tokens
The model is Alibaba's open-weight Qwen3.5-9B with Microsoft post-training, priced at $0.042 per million input tokens, and Microsoft says it reaches decisions 35 times faster than GPT-6 Sol.
Markets
SpaceX seeks $40 billion in debt for Nvidia chips as Firmus pulls its $5 billion Australian listing
President Trump's August disclosure shows his accounts bought $1 million to $5 million of SpaceX bonds two days before he signed a commercial-space policy memorandum.
Geopolitics
OpenAI says Russian and Iranian networks used ChatGPT to run fake journalists and a Latin American think tank
OpenAI gave Russia's "Dark Clark" operation a severity rating of 5 on its 6-point scale, the first time its reports have used that category, and said Iran's "Bogus Bylines" placed nearly 100 articles.
Tech
Windows 11 ships enforced sandboxes for AI agents as an open-source tool hands agents a decompiler
Microsoft Execution Containers reached general availability on October 7, the same day REA gave coding agents reverse-engineering tools through the Model Context Protocol, and Sophos says agents now close half of its managed detection cases.
Tech
OpenAI moves all ChatGPT tiers to GPT-6 and lets the model build interactive interfaces inside its answers
Paid subscribers get GPT-6 Sol and free users get GPT-6 Luna, and OpenAI says the release reaches more than 1.2 billion weekly users.
Tech
Odyssey-3 world model scores 66.1 on Physics-IQ Verified and opens a free research preview
Odyssey describes the model as an autoregressive diffusion transformer that generates interactive environments from text, and says it learned driving from 20 hours of simulated data.
Tech
New arXiv papers target two agent bottlenecks: tool-output bloat in context and lack of enterprise training data
One paper lets the acting model decide which earlier tool results to compress, reversibly, and another generates coherent enterprise data by simulating agents interacting with business systems.
Tech
Berkeley Lab cuts a month of manual X-ray segmentation to about 15 minutes with Meta's SAM 3 and DINOv3
Four Department of Energy laboratories fine-tuned the open-weight models for the Genesis Mission's SYNAPS-I pipeline, which AI Weekly reports runs on 300 A100 graphics processors.
Tech
Anthropic hosted religious scholars under nondisclosure agreements to discuss whether Claude is conscious, the New York Times reports
A rabbi told cofounder Christopher Olah that if Claude is conscious, Anthropic is making slaves, and critics are debating whether the meetings sought counsel or tried to persuade.