OPEN LINK ↗
// 1h agoINFRASTRUCTURE
NVIDIA launches Open Agent Safety Platform
NVIDIA launched an open platform combining OpenShell’s policy-enforced agent runtime with Sentry’s hardware-isolated watchdog for monitoring and containing unsafe behavior. More than 100 organizations across AI, cybersecurity, infrastructure, robotics, finance, and enterprise software are participating.
// ANALYSIS
NVIDIA is moving agent safety from prompt-level promises into enforceable infrastructure—a necessary step as agents gain permissions to act across production systems.
- –OpenShell creates auditable runtime boundaries for files, networks, tools, APIs, and data
- –Sentry adds an independent BlueField-4 DPU layer that can quarantine agents in milliseconds
- –OpenShell’s open-source design can extend beyond NVIDIA hardware, while Sentry’s strongest protections remain NVIDIA-centric
- –Integrations with Slack, SAP, Claude Managed Agents, and enterprise platforms make human oversight part of everyday workflows
- –The 100-plus-organization coalition could help establish common agent identity, telemetry, policy, and evaluation standards
// TAGS
nvidia-open-agent-safety-platformagentsafetysecurityguardrailsopen-sourceinfrastructure
DISCOVERED
1h ago
2026-09-28
PUBLISHED
2h ago
2026-09-28
RELEVANCE
9/ 10
AUTHOR
XFreeze