Five minutes. What happened, and what to do about it. All stories from 20 October 2026
Microsoft extends Project Perception to treat AI agents as insiders, with a permission set up per task and a risk score that changes
Microsoft's design rule for agents: start with no permissions, give each its own identity, and tighten access when its behaviour changes.
NVIDIA releases Nemotron 3.5 Super VL and two Bolt models, open-weight agent models that range from servers to a 24GB PC
NVIDIA released three open models on 15 October. Nemotron 3.5 Super VL has 120 billion parameters, reads images, web pages and PDFs, and NVIDIA says it is five times faster than the next best model in its class. Nemotron 3.5 Bolt 74B-A9B uses 2-bit quantization to fit in 24GB of memory on RTX and DGX Spark machines, so agents can research, write and debug code locally. A 4GB Bolt E4B for voice and conversational agents follows on 23 October. Super VL and Bolt 74B-A9B are available now.
- If you want AI that keeps customer files on your own machine, ask your IT supplier whether a PC with 24GB of graphics memory could run a local agent for one repeat task.
- Pick one document job, such as reading supplier invoices, and list what it would need to read before you test any local model on it.
Source: NVIDIA Blog. This is our short summary of their reporting. The paragraphs marked "Our read" and the things to do are SeedFoundry's opinion. The summary is drafted with AI tools. Image: NVIDIA Blog, from its own announcement. How we source and credit
These are short summaries of other people's reporting, with our own commentary. Each story names its source and links to the original. How we source and credit