Find notable cyber news and cases, enriched with sources, timelines, and signals.
Home Watchlists Beta Browse News Feed Stats About Contact

Nvidia Open Agent Safety Platform launch adds OpenShell sandboxing and Sentry watchdog enforcement

Security Tool/Service
First reported
Last updated
Happening score
H score 20
2 unique sources, 2 articles

Summary

Hide ▲

Nvidia launched Open Agent Safety Platform, adding sandboxing, policy enforcement, and hardware-backed monitoring for AI agents that can drift outside intended boundaries. The release matters because it gives operators a concrete control stack for containing agent behavior from testing through deployment.

Related Happenings

Read-only and scoped permission enforcement for AI agent tool calls

Defensive Guidance
H score16 First: 09.10.2026 17:01 Last: 09.10.2026 17:01 Sources 1

About this happening: Organizations are being urged to enforce read-only and scoped credentials at AI agent tool calls so agents cannot overreach through prompt injection, mistaken assumpti...

Open Happening

OpenAI Astra reaches Critical cybersecurity threshold with Daybreak Blue tester access

Security Tool/Service
H score18 First: 02.09.2026 21:27 Last: 02.09.2026 21:27 Sources 1

About this happening: OpenAI has moved Astra into a controlled tester-access phase after classifying it at the Critical cybersecurity capability threshold, raising the stakes for advanced AI-as...

Open Happening

NVIDIA Ampere-class GPUs GPUThor Rowhammer ECC bypass privilege-escalation flaw

Vulnerability
H score26 First: 26.08.2026 21:48 Last: 26.08.2026 21:48 Sources 1

About this happening: GPUThor is a Rowhammer vulnerability affecting NVIDIA workstation GPUs with GDDR6 memory that can defeat ECC and enable denial-of-service plus host root pr...

Open Happening

TRACE open standard for AI runtime evidence and hardware-attested AI governance records

Security Tool/Service
H score11 First: 26.08.2026 12:10 Last: 26.08.2026 12:10 Sources 1

About this happening: TRACE introduces a new way to prove AI runtime evidence, giving organizations a verifiable security control for what agents actually did, which policies applied, and which...

Open Happening

NCSC interim sandboxing guidance for autonomous AI agents

Defensive Guidance
H score16 First: 20.08.2026 15:45 Last: 20.08.2026 15:45 Sources 1

About this happening: NCSC issued interim guidance for autonomous AI agents, recommending sandboxing, human oversight, and tightly controlled access to reduce the risk of unintended...

Open Happening

Timeline

  1. 28.09.2026 13:27 3 articles · 13d ago

    Nvidia launches Open Agent Safety Platform for AI agent containment

    Initial Disclosure

    Nvidia announced the Open Agent Safety Platform, combining the OpenShell open source runtime with Sentry on BlueField-4 DPUs to sandbox agents, enforce policy, and monitor behavior outside the host. The release says OpenShell is broadly available at version 0.1.0 and that Sentry can quarantine agents that move outside their software boundary.

    Show sources
    Open in new tab