Check Point joins Nvidia design to secure AI factories

Check Point Software has added its AI Cloud Protect product to Nvidia's Enterprise AI Factory validated design, positioning the cyber security firm inside a reference architecture that targets organisations running AI workloads in dedicated data centre environments.

The companies described the move as a step focused on AI runtime security for "AI factories", a term used by Nvidia and others to describe purpose-built infrastructure for training and running AI models. The validated design sits within Nvidia's broader programme of reference configurations for enterprise AI deployments.

The announcement comes as businesses increase spending on AI tooling and infrastructure, while security teams assess risks that range from attacks on AI pipelines to manipulation of prompts and model behaviour. The advisory cited third-party research that suggests such incidents already occur in production environments.

The statement referenced a Gartner report that said 32% of organisations had experienced an AI attack involving prompt manipulation and 29% had faced attacks on their generative AI infrastructure in the past year. It also cited a survey from Lakera, which found that 19% of organisations described their generative AI security posture as "highly confident", while 49% reported high concern about vulnerabilities.

Validated Design

Check Point said AI Cloud Protect is now part of the Nvidia Enterprise AI Factory validated design for AI runtime cyber security. The product is positioned as a security layer that works across infrastructure where AI workloads run.

Check Point also said its software is "validated on NVIDIA RTX PRO Servers". It said the combination is aimed at securing AI factories at scale. The advisory claimed there is "no negative impact to AI system performance".

In the same statement, Check Point described AI factories as "the new class of purpose-built data centers for AI." That framing reflects a wider shift among vendors that market AI hardware and software stacks as integrated platforms rather than standalone components.

Network Integration

Check Point linked its approach to Nvidia's BlueField platform. It said the integration "tackles cyber threats and vulnerabilities" with real-time monitoring and isolation between AI workloads. It also said the integration provides "deep visibility and control over AI data".

The advisory stated that the integrated AI Cloud Protect software delivers real-time network and host security using Nvidia DOCA Argus telemetry and Check Point's "native AI-powered cyber security". It positioned telemetry as a data source that feeds detection and response functions during runtime operations.

Check Point also described a multi-layer model that spans infrastructure, applications and users. It framed this as an "AI supply chain" approach, where security policies cover components that sit on different parts of the AI stack.

Three Layers

For the infrastructure layer, Check Point said AI Cloud Protect runs on Nvidia BlueField and secures AI infrastructure "without consuming precious GPU capacity". The statement again repeated its claim of "zero negative performance impact".

For the application layer, Check Point highlighted CloudGuard Web Application Firewall. It said the product stops AI application threats, including "prompt injection, jailbreaking, and LLM poisoning". It also referenced runtime protection for "LLM inputs, outputs, and all data flows-including retrieval-augmented generation (RAG) and model context protocol (MCP) servers."

Check Point claimed it has a "unique data advantage from Gandalf" and described it as "the world's largest AI red team platform with over 80 million adversarial attack patterns". The company linked this to detection quality and low false positives in its application security layer.

For the user layer, Check Point described GenAI Protect as a tool that governs employee AI usage and prevents sensitive data leakage. It also said the product provides visibility into AI tools employees use and generates audit trails for compliance purposes.

Firewall Controls

The advisory also outlined network-layer controls via Check Point firewalls, naming Quantum and CloudGuard Network Security. It listed features such as "Auto-detect MCP servers and traffic (Model Context Protocol)" and policy controls that permit or block access to AI applications.

The company also referenced "Full control over Agentic AI, MCP servers, and shadow Gen AI apps". These terms have become more common as organisations encounter employee-led experimentation with AI tooling outside formal procurement and governance processes.

As organisations deploy AI systems across business functions, suppliers increasingly position security tools around runtime monitoring, application-level protections, and governance for employee usage, alongside traditional network and infrastructure controls.