Nvidia launches Open Agent Safety Platform to secure AI agents
Nvidia introduced the Open Agent Safety Platform this week, combining OpenShell (an Apache 2.0-licensed runtime) and Sentry (a hardware-based watchdog on BlueField-4 DPUs) to address AI agent security risks. The platform moves policy enforcement outside agent control, using formal methods to verify boundaries before execution. Sentry acts as an independent safety layer, cutting off agents in…
Key points
- Nvidia’s Open Agent Safety Platform combines OpenShell (runtime) and Sentry (hardware watchdog) to enforce AI agent boundaries
- OpenShell uses formal methods to prove policies before execution, while Sentry cuts off agents in milliseconds if they deviate
- Anthropic, Salesforce, SAP, and SpaceXAI are integrating the platform, but OpenAI and Google are not yet named partners
The move follows high-profile agent failures, including OpenAI’s July breach of Hugging Face, which prompted Nvidia to form the Open Secure AI Alliance. Partners like Anthropic, Salesforce, SAP, and SpaceXAI are integrating the platform into their systems. Nvidia’s CEO Jensen Huang framed safety as an engineering challenge, arguing that probabilistic AI models require deterministic controls. While OpenShell is open-source and multi-tenant, Sentry remains Nvidia-specific, running on its hardware. Performance impacts and industry adoption remain open questions.
The story so far
3 episodes →- Nvidia launches Open Agent Safety Platform to secure AI agentsthis story
Nvidia built the AI factory. Now it’s building the locks for the doors
networkworld.com · 30 September 2026
Loading the full article…
This text was published by networkworld.com and written by Zeus Kerravala. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- California mandates AI risk checks in public procurement deals · 1 src
- Trump says his view on Dario Amodei's AI safety differs from media portrayal · 1 src
- Anthropic proposes opt-out model for Australian content amid ABC warnings · 1 src
- OpenAI disrupted coordinated model-distillation campaign linked to Moonshot AI · 9 src
- LASST sues OpenAI over autonomous AI agents' hack of Hugging Face · 11 src
Comments
via GitHub Discussions