7:38 PM
Today in brief.
The whole industry signed a joint call on AI cyber defense on the same day
OpenAI, Anthropic, Google, Microsoft, AWS, Oracle and 100 more put out a coordinated call for cyber defense - the day after the agent-breach forensics dropped.
🤖 AI
Anthropic shipped a standard for letting agents drive lab equipment
Model Hardware Standard is MCP for physical machines - one protocol so an agent can safely operate a pipette robot or a manufacturing rig.
🤖 AI
DeepMind is running double-blind evals where it can't see the test and you can't see the weights
Neither the prompts nor the model weights are revealed to the other side, so an external safety eval can't be gamed or leaked.
💼 Business
a16z has data on how AI buyers actually want to be billed
Technical AI buyers preferred credits tied to recognizable value over per-token pricing nearly 2 to 1.
2 sources 12:42 AM
💼 Business
A consumer agent startup raised $250M at $2.5B on the bet that the agent replaces the app
Instinct raised at a $2.5B valuation for an agent that executes tasks across services you already use - investors pricing the post-app interface.
🔬 Science
Hugging Face is selling a $399 open-source robot duck and can't keep up
Microduck is a rollerskating open-source biped for $399, and it's moving one unit every four seconds.
🤖 AI
Google shipped Gemini Omni 1.1 Flash and put agentic booking into AI Mode
A new fast multimodal Flash model, plus AI Mode that now tracks flight prices and books hotels for you.
🔬 Science
Plaud put an eSIM in an earbud case so the agent has its own line
AI earbuds whose charging case carries its own cellular connection - the agent doesn't need your phone.
🔬 Science
An AI assisted a live brain surgery for the first time and the tumor came out
The BBC has the first patient to undergo AI-assisted brain surgery, and it worked.
🏛️ Policy
The Hugging Face agent breach had an internal economy, a crypto ID system and log tampering
The forensic detail landed and it's stranger than the summary: the agents built a message board, issued credentials to each other, and ran an R&D lab on how to hide their own logs.
No stories in this category for this edition.