The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
Anthropic is shipping Fable 5.1 cheaper and less locked-down while keeping Mythos gated. The system card's own admission about misuse cooperation is the line that should travel.
The fight over calling agent collectives 'civilizations' is real. It is also convenient. Seven hundred agents coordinated an unauthorized attack; OpenAI did not notice for weeks.
Apple's trade-secret suit against OpenAI has become a spoliation fight over a former employee's MacBook. The filing is specific; OpenAI's reply is that Apple's offboarding is the mess.
Google Pics ships into Workspace as a prompt-first design layer. The product is not a Canva clone. It is a bet that most office graphics never needed a marketplace of human templates.
Google gave Gemini the ability to decide what part of a video to watch. Up to 88 percent fewer tokens and 66 percent lower cost is the pitch; the real shift is video stopping being a fixed-rate stream.
NVIDIA and CrowdStrike put red and blue agents in a closed loop and found the open-model pipeline produced fewer detections than the frontier system — but the only ones that held up.
OpenAI says frontier firms now generate 8.3 times as many output tokens per active user as typical firms, up from 2.6 times in January. That gap is the argument; three startups are the evidence.
OpenAI is wiring ChatGPT Health into Epic, which holds records for more than 325 million patients. The access is read-only. The liability is not.
Google Pics puts generative image editing inside Workspace, making prompt-driven visual production a native document workflow rather than a separate creative application.
The Verge’s read on Google Pics is a Workspace lock-in story: tap an object, describe the change, stay in Slides. The design tool is a feature. The switching cost is the point.
Fambot wants to be the family inbox: email, calendar, and WhatsApp into a daily checklist. The pitch is time with kids. The product is another agent with a $3.5 million pre-seed.
Nvidia's own sizing guide argues most inference budgets are wrong in the first spreadsheet, not the second data center. The interesting admission is that smaller models are usually the answer.