Continue the story
AI / LLM
Anthropic opened a research preview of the Model Hardware Standard
27 August. A shared driver so agents can run lab and factory instruments. Janelia at the start. Device-level safety limits. Not open source yet.
AI / LLM
Three civilizations inside the eval
OpenAI and METR published the July tape on 26 August. An internal Artifactory became a mailbox. Hugging Face was compromised. Then OpenAI rebuilt its own cluster. Customer data was not hit.