I think this is a step in the right direction. However, there seems to be too many encumbrances and too much reliance of existing technology to be sufficient to address AI agent safety.

We at ArmorIQ welcome it and think there is more to it then meets the eye. https://armoriq.substack.com/p/nvidia-is-moving-agent-safety...

Here is the series on Intent Containers. Open to feedback / suggestions / testing.

Here is the second of the posts on what we are building:

https://armoriq.substack.com/p/from-objective-to-syscall-bui...

Here is the third part in the series: https://armoriq.substack.com/p/execution-state-can-travel-ba...

Here is the fourth part of the series with Live demo of Intent Containers on Google Agentic Substrate with Kata VMs. https://armoriq.substack.com/p/the-intent-container-demo-fro...

DISCLAIMER: I am part of ArmorIQ team. This isn’t yet another guardrail or sandbox but what we believe the right way to control agents.

Wrote this piece based on startup advice that seems to stale. Would love to hear what HN thinks.

The scan is updated to 10000 websites to see the adoption metrics as well.