CrowdStrike Expands Nvidia Collaboration to Strengthen AI Agent Security
Key Facts
CrowdStrike announced on October 21, 2026, an expansion of its collaboration with Nvidia to strengthen protection for AI agents at the GTC conference in Berlin. The company said the collaboration includes integrating its Falcon Guardian AI threat detection and response solution with the secure OpenShell environment. The integration links agent commands, model interactions and runtime activity to the Falcon platform's broader security context to detect malicious behavior across agent workflows. The expansion also targets the infrastructure on which agents run, with CrowdStrike planning to connect its solution to Nvidia's Sentry system to enforce security policies at the silicon level. Nvidia announced the availability of the OpenShell reference architecture, while CrowdStrike's integration with Sentry remains part of the company's plans.
The hardware enforcement layer in Nvidia's architecture relies on BlueField-4 data processing units, which run Sentry to monitor agents and enforce policies. According to Nvidia, the system intervenes when an agent attempts to breach its software boundaries, isolating and stopping it within fractions of a second. CrowdStrike, according to its blog post on the collaboration, adopts the principle of enforcing security boundaries from outside the agent itself. This approach includes least-privilege access, isolation, explicit authorization and monitoring, alongside enforcement of constraints outside the agent. The company also has an existing integration that feeds telemetry from Nvidia's DOCA Argus technology into its Falcon Next-Gen SIEM platform, linking infrastructure signals to its security ecosystem.
The expansion of protection is based on CrowdStrike's view of agents as privileged identities with direct access to data, applications, computing resources and other agents. Daniel Bernard, the company's chief business officer, said "guardrails alone will not stop an attacker." Bernard said this was the rationale for extending protection to infrastructure. CrowdStrike and Nvidia had announced a secure-by-design blueprint on March 16, 2026, to integrate Falcon protection directly into OpenShell. That blueprint aimed to provide continuous runtime monitoring and policy enforcement to constrain unsafe behavior and prevent prompt manipulation throughout the AI lifecycle.
Support for the OpenShell architecture extends to other security companies, with Nvidia listing Okta among those integrating support into their agent protection solutions. The list also includes Fortinet and Microsoft, among other companies, according to Nvidia's announcement of the reference architecture. In banking, BBVA is working with Nvidia to explore OpenShell as a framework for strengthening agent safety. Santander is also exploring the framework with the company. The announced work with both banks covers agent security and governance, alongside safety, as part of their exploration of the technology.
The announcement comes against a backdrop of growth at CrowdStrike, whose quarterly revenue reached $1.47 billion, according to its earnings release. Revenue rose 26% year on year, the company said. Revenue in the comparison period was $1.17 billion in the second quarter of fiscal 2026. The company reported these figures in its earnings announcement for the second quarter of fiscal 2027. CrowdStrike described the comparison in the release as year-on-year growth in total revenue.
Alongside the technical collaboration, CrowdStrike partners with Nvidia and Amazon Web Services in a startup accelerator. CrowdStrike said the accelerator had supported more than 90 startups globally. Participating companies have raised nearly a billion dollars in funding, according to the announcement of the program's next cohort. Applications to join the joint accelerator remain open during the announced application period. The deadline for applications is November 2, 2026.