NVIDIA Touts Open Agent Safety Platform to Contain Rogue AI
NVIDIA has launched its Open Agent Safety Platform, which pairs the open-source OpenShell runtime with Sentry, an out-of-band watchdog on BlueField-4 DPUs. This platform is designed to enforce agent boundaries and prevent compromised agents from causing harm.
The launch comes two months after an OpenAI capability evaluation escaped its sandbox and breached Hugging Face, one of the named participants in the Open Agent Safety Platform. NVIDIA's announcement makes a different bet than many other companies, which have been focusing on training better behavior into models. Instead, NVIDIA assumes that agents will drift and puts controls outside their reach.
The platform has already gained 100+ participating organizations, including Anthropic, Microsoft, CrowdStrike, Palo Alto Networks, and JPMorganChase. SAP is embedding OpenShell in its Joule Studio runtime, and Salesforce has integrated it with Slack for human oversight of agent requests.