Anthropic and Nvidia jointly launched the Open Agent Safety Platform on September 28, 2026. The platform combines Anthropic’s Claude Managed Agents architecture with Nvidia’s hardware-level controls, notably the OpenShell and Sentry technologies running on BlueField-4 Data Processing Units. Microsoft, Cisco and CrowdStrike are among the partner organizations, while OpenAI is notably absent from the list. The architecture physically isolates an AI agent’s reasoning loop from its execution sandboxes, creating a layered containment system with hardware-level monitoring that is harder to circumvent. The two companies had previously collaborated on Project Glasswing, launched in April 2026, focused on building defenses against AI-enhanced cyber threats.
Source: Read the original article
