Security Integrated at Launch on Major AI Platform
For the first time, security has been shipped at the launch of a major AI platform release, rather than being added later. At Nvidia GTC this week, five security vendors have announced protection for Nvidia’s agentic AI stack, with four having active deployments and one validated early integration.
The urgency stems from the rapidly evolving threat landscape, with 48% of cybersecurity professionals ranking agentic AI as the top attack vector heading into 2026. Only 29% of organizations feel fully prepared to deploy these technologies securely. Machine identities now outnumber human employees by 82 to 1 in the average enterprise. IBM’s 2026 X-Force Threat Intelligence Index reported a 44% surge in attacks exploiting public-facing applications, driven by AI-enabled vulnerability scanning.
Nvidia CEO Jensen Huang emphasized the need for security, stating, “Agentic systems in the corporate network can access sensitive information, execute code, and communicate externally. Obviously, this can’t possibly be allowed.”
Unified Threat Model and Collaborations
Nvidia has defined a unified threat model that accommodates the unique strengths of five different vendors. Notable collaborations include Google, Microsoft Security, and TrendAI as Nvidia OpenShell security collaborators.
The article maps out the five vendors with embargoed GTC announcements, verifiable deployment commitments, and an analyst-synthesized reference architecture.
Five-Layer Governance Framework
The framework comprises five governance layers, each with specific questions that vendors should address. The governance layers include Agent Decisions, Local Execution, Cloud Ops, Identity, and Supply Chain.
Security leaders are advised to evaluate different vendors for specific governance layers, such as CrowdStrike for agent decisions and identity, Palo Alto Networks for cloud runtime, JFrog for supply chain provenance, Cisco for prompt-layer inspection, and WWT for pre-production validation.
The article also highlights the importance of auditing governance layers to ensure comprehensive security coverage and readiness for deploying agentic AI technologies securely.
Vendor Stack and Deployment
Each of the five vendors occupies a distinct enforcement point within the security stack. CrowdStrike, Palo Alto Networks, JFrog, WWT, and Cisco offer complementary security capabilities that collectively address different aspects of agentic AI security.
Several enterprises have already deployed the CrowdStrike-Nvidia stack to power their Agentic SOC services, with positive feedback on the enhanced security capabilities.
Challenges and Recommendations
The article also highlights three key challenges that the current vendor stack does not fully cover, including agent-to-agent trust, memory integrity, and registry-to-runtime provenance. Security leaders are advised to address these gaps before scaling their agentic AI deployments.
Before the next board meeting, CISOs are encouraged to conduct a comprehensive audit of autonomous agents against the five-layer governance framework, identify unanswered questions, pressure-test open gaps with vendors, and establish a robust oversight model to ensure effective governance of agentic AI technologies.
Overall, the article emphasizes the importance of integrating security measures at the outset of AI platform releases and collaborating with multiple vendors to create a comprehensive security architecture for agentic AI deployments.
