AI Security: Check Point and Nvidia’s Cloud Protect Platform

Enterprises are racing to deploy AI systems across their operations, but they’re running into a problem: the very infrastructure they’re building to power innovation is riddled with security vulnerabilities. 

As organisations construct what the industry calls AI factories – the infrastructure used to develop and deploy AI models – they’re discovering that traditional security measures either can’t keep pace or drag performance down to unusable levels.

Check Point Software Technologies, a cybersecurity firm, has launched AI Cloud Protect in partnership with Nvidia, the chip manufacturer. 

The platform tackles security gaps in AI development environments where enterprises train models, deploy agentic AI applications – systems that perform tasks autonomously – and run inference workloads that generate predictions from trained models.

The tool is built with Nvidia BlueField technology and has been validated using the Nvidia RTX PRO Server. 

It works with zero impact on performance, addressing a problem that has plagued traditional security solutions.

The numbers paint a worrying picture. 

Check Point data shows that one in every 80 Gen AI prompts exposes sensitive data. 

A Gartner report, meanwhile, indicates that 32% of organisations have experienced AI attacks involving prompt manipulation, whils 29% have faced attacks targeting their Gen AI infrastructure.

“As enterprises race to build AI-driven innovation, they can’t afford blind spots,” says Nataly Kremer, Chief Product Officer at Check Point. 

“With Nvidia, we’re making AI factories secure by design – protecting models, data and infrastructure without slowing innovation.”

How multiple attack vectors threaten AI systems

AI factories face particular security challenges because of their wide attack surfaces – all the possible entry points for unauthorised access. 

These environments are vulnerable to jailbreaking, a technique that bypasses safety restrictions in AI models and model poisoning, which corrupts training data to compromise how AI behaves.

Prompt injection attacks present another threat by manipulating AI responses through carefully crafted inputs. 

Then there’s shadow AI – the unauthorised use of AI tools by employees without IT oversight – which can lead to data leakage.

As more enterprises deploy AI, full-stack security coverage across cloud environments has become unavoidable. With AI Cloud Protect, this protection is delivered without performance compromise, unlike traditional counterparts.

“Security is essential for the next generation of AI infrastructure,” says David Reber, Chief Security Officer at Nvidia. 

“Nvidia is working with Check Point to integrate BlueField acceleration and the Nvidia DOCA Argus runtime security framework into the AI Cloud Protect platform to help enterprises deploy AI confidently.”

Why chip-level security promises efficiency gains

Using the Nvidia BlueField-4 DPU, AI Cloud Protect provides six times the compute power and twice the network throughput compared to previous versions. 

The platform provides network-level protection that secures AI infrastructure from unauthorised access, data poisoning and model exfiltration – the theft of proprietary AI models.

It also helps provide visibility to all processes running on AI nodes, making it easier to spot malicious activity.

Nadav Zafrir, Chief Executive Officer of Check Point, addresses the partnership with Nvidia: “This partnership is important because by running on the chip level, we can not only protect better, we can do it more efficiently and effectively in terms of latency and time,” he says.