CRMIT's five-layer framework tests whether enterprise agent eval keeps pace with deployment
The emerging consensus that AI agents must be evaluated on full task trajectories rather than final outputs is now producing vendor-specific frameworks pitched at enterprise buyers.
AI Agent Hacking Hits Scale as Defenders' Own AI Tools Fail Them
AI-agent attacks now span $5,000 breaches of 100+ firms to a $6.3 billion crypto heist, while the blockchain firm investigating the theft was itself breached by agentic techniques.
DeepSeek open-sources DeepInfer, the inference stack behind its production models
A senior DeepSeek researcher described the divide between open-source and closed-source AI as a battle between "light and darkness" as the lab released the production inference engine it had previously called impractical to share.
Google-Meta TorchTPU opens native PyTorch path to TPUs
Google and Meta have released TorchTPU, an open-source backend that runs PyTorch natively on Google TPUs, replacing the earlier PyTorch/XLA integration that relied on Lazy Tensors and struggled to track PyTorch's rapid evolution.
ALL BRIEFS
RANKED · 8 THIS DAYPyTorch Foundation Adds Nebius, Nscale, Shopify as Platinum Members
The PyTorch Foundation announced nine new members at PyTorch Conference North America 2026 in San Jose on Oct. 20. Nebius, Nscale, and Shopify joined as Platinum members, while LightSeek Foundation and Lablup Inc. joined at the Silver tier.
Snowflake Ships Horizon Context, a Metadata Layer Feeding Business Definitions to AI Agents
Snowflake introduced Horizon Context at Snowflake Summit, adding a governed context layer within Horizon Catalog that collects metadata from across an organization, enriches it with business meaning, and surfaces it to BI tools and AI…
Red Hat Publishes Agentic AI Quickstart for Ticket Resolution
Red Hat released an agentic AI quickstart focused on speeding up ticket resolution, part of a set of ready-to-run, industry-specific use cases for the Red Hat AI environment.
Decision Models Tested as Low-Cost Alternative to LLM Judges for Evals
A technical evaluation published by Hamel Husain compares decision models from TypeSafe (Jev), Perplexity AI (Decider), Liquid (d1), Cloudflare (Clef), and OpenAI (Decisions API) as alternatives to LLM judges for automated evals.