Tech Thought Leaders

Independent reporting on Kubernetes, cloud-native infrastructure, virtualization, storage, and security. For engineers and platform teams.

Bites

Nvidia Moved the Agent Guard Off the Machine, in a Bite

ByIvan Tarin

Oct 1, 2026
Photo by ThisisEngineering on Unsplash. Source: https://unsplash.com/photos/woman-in-black-shirt-sitting-on-chair-in-front-of-computer-pIdteYOzGFY (Unsplash License).

Nvidia’s answer to rogue AI agents is to stop putting the guard where the agent can already reach it.

  • NVIDIA Open Agent Safety Platform shipped September 28, split into two parts that never share a machine.
  • OpenShell is the open source software boundary. It traces actions and sets what an agent can reach.
  • Sentry is a watchdog on the BlueField-4 DPU. NVIDIA says it quarantines a breakout in milliseconds.
  • That matters because this year’s agent incidents all broke at the application layer, where the agent already had reach.
  • Anthropic runs the same idea without new hardware, by splitting the agent loop from the sandbox.

The catch is that the boundary runs on one vendor’s silicon. Read the full analysis.

By Ivan Tarin

Ivan Tarin is a Principal Product Marketing Manager at SUSE, where he owns go-to-market strategy and positioning for a seven-product cloud-native portfolio spanning Kubernetes, virtualization, storage, security, and observability. A former full-stack developer who shipped production code for enterprise and public-sector clients including U.S. national laboratories, Ivan translates complex infrastructure and AI technology into messaging that lands with developers, platform teams, and enterprise buyers. He has presented at KubeCon, SUSECON, and AWS Developer Week, and is currently pursuing an MS in Artificial Intelligence at the University of Colorado Boulder.

Related Post

Bites

Sovereign AI Capacity Costs $60 Million a Megawatt, in a Bite

Oct 5, 2026 Ivan Tarin
Bites

Intel Raised CPU Prices Again and Memory Is the Reason, in a Bite

Oct 5, 2026 Ivan Tarin
Bites

Goldman Kept the Data Center Number and the Backlash Did Not Dent It, in a Bite

Oct 5, 2026 Ivan Tarin

Leave a Reply

Your email address will not be published. Required fields are marked *

You missed

Analysis Platform Engineering

Kubernetes Node Swap Triples Agent Density When Memory Is the Constraint

October 5, 2026 Ivan Tarin
Analysis

The Cheapest Good Model Costs 2.9 Cents a Task and the Best One Costs 42 Times That

October 5, 2026 Ivan Tarin
Analysis

Two of GitHub’s Three Biggest Projects Are Agent Harnesses, Not Models

October 5, 2026 Ivan Tarin
Analysis

NVIDIA Stopped Selling GPUs. It Sells the Rack, and That Rewrites the Budget Line.

October 5, 2026 Ivan Tarin

Get the next one before it is old news

Independent analysis of cloud-native infrastructure, Kubernetes and data center economics. No vendor spin.