AI agents are already operating across the enterprise. As they become more autonomous in accessing data, using credentials, executing code, and acting across critical systems, the security boundary is changing with them.
The question is no longer whether enterprises will adopt agents. It's how they give agents greater autonomy without giving up control.
Today, NVIDIA introduced the NVIDIA Open Agent Safety Platform, an open source platform and reference design that brings multi-layered security controls across the software, runtime, and infrastructure where agents operate. CrowdStrike and NVIDIA are collaborating on this effort to extend our work together deeper into the agentic stack.
Model safeguards are essential, but they aren't the entire security boundary. When agents operate with identities, credentials, and access to critical systems, enterprises need controls that work where agents run and act.
Agent safety is a cybersecurity problem. It demands a full-stack approach.
That's the approach CrowdStrike and NVIDIA are advancing together. We are empowering AI to transform cyber defense and securing the AI transforming the enterprise.
The Agent Can't Be Its Own Security Boundary
An agent shouldn't be responsible for enforcing its own security boundaries.
We already understand this everywhere else in cybersecurity. We don't rely on applications or workloads to determine whether they've been compromised. We establish independent controls around them to govern access, monitor behavior, and enforce policy.
Agents are no different. Models and agent harnesses can guide behavior. But once an agent starts interacting with real systems, we need least privilege, isolation, explicit authorization, observability, and enforcement outside the agent itself. An agent should be treated as a privileged identity with permissions bounded by policy, not expanded simply because it determines it needs more access.
These aren't new security principles. The interesting problem is applying them to autonomous systems operating at machine speed.
New Control Points and Additional Data to Secure AI
NVIDIA Open Agent Safety Platform brings together security across multiple layers of the agentic stack. At the software layer, technologies like NVIDIA OpenShell provide an open source secure runtime that governs how agents execute, what they can access, and where inference occurs. NVIDIA Sentry adds an independent watchdog running on NVIDIA BlueField-4 DPUs that uses NVIDIA DOCA for out-of-band, in-silicon monitoring of agent activity and security policy enforcement. It provides attested telemetry, verifies agent identity, and enforces Zero Trust access to data, tools, APIs, and services. During agent testing and evaluation, NVIDIA Sentry can quarantine agents in milliseconds. NVIDIA Vera CPUs provide high-performance compute for agentic workloads.
These technologies establish independent security boundaries closer to where agents run and act. The CrowdStrike Falcon® platform connects that activity to the broader security context of the enterprise.
Broader Context Helps Turn Signals into Protection
Consider something as simple as an agent accessing a file, calling an API, or making a network connection. None of those actions is inherently malicious. What matters is the context. Which agent is acting? Under what identity? From which workload? What data is it accessing? Is the behavior expected? What else is happening across the environment?
The new control points and signals across NVIDIA Open Agent Safety Platform become even more valuable when connected to the broader business security context. The Falcon platform provides that context by bringing together security across endpoint, identity, cloud, data, and AI, combining rich security telemetry with adversary intelligence and years of frontline expertise.
This gives defenders and autonomous defense agents the context to understand agent activity alongside everything else happening across the enterprise, and the ability to detect and stop malicious behavior from a unified security platform.
Building on Our Collaboration with NVIDIA
With the recent launch of CrowdStrike Falcon® Guardian, we have extended protection across the AI attack surface, including runtime protection for AI agents. Earlier this year, we unveiled a Secure-by-Design AI Blueprint with NVIDIA, bringing Falcon protection together with NVIDIA OpenShell to embed governance, visibility, and control into the AI agent stack. At the infrastructure layer, our existing integration of NVIDIA DOCA Argus telemetry with CrowdStrike Falcon® Next-Gen SIEM brings infrastructure-level telemetry into the Falcon platform.
Now we are advancing this work across the agentic stack by bringing independent controls closer to where agents operate with the enterprise-wide visibility, intelligence, and protection of the Falcon platform. That's how we turn agent activity into security context, and security context into protection at machine speed.
It Takes a Crowd to Secure AI
The next phase of AI will be defined by increasingly autonomous systems operating across the enterprise, making decisions, using tools, and acting on our behalf.
A single layer cannot secure that future. Models and harnesses can guide behavior. Runtimes can establish boundaries. Hardware can provide independent enforcement. Cybersecurity has to connect these layers to what is happening across the enterprise with the outcome of securing the inputs and the outputs.
What’s more, no single technology or company will secure that future. It will take an ecosystem with the best models, infrastructure, runtimes, and security technologies working together. That's the vision behind our work with NVIDIA, and why we'll continue extending security across the AI stack while bringing that context together through the Falcon platform.
Machine-speed threats require machine-speed defense. The frontier will continue to move. Our job in cybersecurity is to make sure it moves securely.
Additional Resources
- Read more about collaboration between CrowdStrike and NVIDIA in this blog: CrowdStrike Scales AI-Native Agents Across Falcon Exposure Management with NVIDIA
- Read this press release: CrowdStrike Unveils Secure-by-Design AI Blueprint for AI Agents Built with NVIDIA
- Learn more about how CrowdStrike is helping to secure the AI era: CrowdStrike Joins the Open Secure AI Alliance to Advance AI Safety and Security
- Read about CrowdStrike SafeMind, the first agentic system for defenders, built with NVIDIA Nemotron: CrowdStrike Launches Frontier Models for Cybersecurity, Created with NVIDIA