Check Point Software Technologies has unveiled the AI Factory Security Architecture Blueprint, a fully tested reference framework designed to secure private AI infrastructure across every layer—from hardware GPUs to LLM inference endpoints—leveraging deep integration with NVIDIA’s BlueField Data Processing Units (DPUs). Released amid explosive growth in sovereign AI data centers, the blueprint addresses the unique vulnerabilities of high-performance environments combining massive GPU clusters, distributed training pipelines, data lakes, and real-time APIs.
The AI Data Center Threat Landscape
Unlike traditional data centers, AI factories expose novel attack surfaces: training data poisoning, model inversion/theft, prompt injection at inference layers, Kubernetes namespace lateral movement, and supply chain compromises via OSS dependencies. “AI infrastructure has become one of the most valuable and vulnerable assets in the enterprise,” said Nataly Kremer, Check Point Chief Product Officer. “This blueprint protects those investments from the ground up.”
Four-Tier Zero Trust Architecture
- Perimeter Layer: Check Point Maestro Hyperscale Orchestrator enforces Zero Trust Network Access (ZTNA) and scalable policies for all north-south traffic into the AI fabric.
- Application & LLM Layer: Purpose-built AI Agent Security safeguards inference APIs and LLM endpoints against prompt injection, data exfiltration, and RCE—threats bypassing conventional firewalls. Available across cloud, virtual, and physical form factors.
- AI Infrastructure Layer: NVIDIA BlueField DPUs via DOCA platform embed hardware-accelerated threat prevention, offloading security from CPUs/GPUs for tenant isolation and inline inspection.
- Workload & Container Layer: Microsegmentation controls east-west traffic in Kubernetes, containing breaches within namespaces.
Secure-by-Design Alignment
Aligned with CISA’s Secure by Design principles, the blueprint embeds security natively rather than retrofitting production systems. Every API call, user, and service undergoes continuous authentication and validation under Zero Trust.
Check Point’s approach positions AI factories for safe scaling, protecting multi-billion IP investments as enterprises prioritize on-prem sovereignty and cost control over public cloud risks.
