Crowdstrike Falcon is now feeding runtime defense in Nvidia’s LLMS

Generative AI adoption has risen by 187Calculation in the past two years. But at the same time only 43%Creating an SIGni gap in readyness as ai -capture surfaces expand quickly.

More than 70% In the past year alone, companies experienced at least one AI-related infringement, with generative models now the primary target, according to recent Without institute Findings.

Attacks on AI infrastructure have a stunning attacks by the state 218% year-on-year, ash Crowdstrike’s 2025 Global Threat Report reveals.

The harsh reality is clear for cisos, safety and soc -leaders. Implementing new AI models on a scale expands the attack surfaces of their companies and Cisos that speak on condition of anonymity have said that the traditional security tactics, strategies and technologies of Venturebeat have been told to keep pace. The cyber security industry has reached a critical bending point: securing generative AI requires more than bolt-on tools; It requires a complete architectural shift

Fortunately, Crowdstrike also offers a new solution: on 11 June at the GTC Paris event of NVIDIA, the security company announced that the Falcon Cloud Security had embedded directly within Nvidia’s Universal LLM NIM. The integration protects more than 100,000 LLM implementations on Enterprise scale in the hybrid and multi-cloud environments of Nvidia.

Crowdstrike’s strategic reaction

Crowdstrike CEO George Kurtz has recorded the urgency in a recent interview with Venturebeat: “Security cannot be screwed up; it must be intrinsic. A significant part of our strategy has always been to use security data as an important element of our core infrastructure. You cannot protect AI without data and visibility at the Lagen.”

“Nvidia’s Nemo Safety offers a framework for evaluating AI-risk. Crowdstrike’s threat information improves that framework by security and operating teams to build crash barriers around rising AI exploites, chief, business, Berdike’s business, Berdike, in the Wildike, business, in the wild, business, one-in-the-game, Bernike, one-in-the-game. Recent interview with Venturebeat.

Kurtz strengthened this strategic vision to BARron’sSays clearly: “Generative AI helps us to bend time. With embedded, telemetry driven by telemetry, we identify and neutralize threats with machine speed, which means that infringements probably stop six times faster than traditional methods.”

Bernard benadrukte het belang en zei: “CrowdStrike was een pionier in AI-native cybersecurity en we bepalen hoe AI is beveiligd in de Lifecycle van de softwareontwikkeling. Deze nieuwste samenwerking met NVIDIA brengt ons leiderschap naar de voorgrond van cloudgebaseerde AI, waar LLMS wordt gedeeld, run, run, en scaled, en scaled, en scaled, en Scaled, run, and scaled, and scaled, and scaled, and scaled, and scaled, and scaled, and scaled and scaled, and scaled, and scaled, and scaled, and scaled, and scaled, and scaled, and scaled, and scaled, and, and, and, and, and, and,, and Scaled, and Scaled, and Scaled, and Scale. “

Crowdstrike encloses Falcon Security directly into the AI ​​infrastructure of Nvidia

By bed directly in Nvidia’s LLM NIM Microservices by Falcon Cloud Security, Crowdstrike Runtime -Protection produces where threats actually occur: within the AI ​​pipeline itself.

“AI is not an independent initiative -it is embedded in the company. Unlike many cloud protection suppliers that are demolished at AI options, we have built AI protection directly in the Falcon platform. This allows us to offer protection that is united in the cloud, which is more and more of the crucial,” “” “” “” “” “” “” “” “” “” “” “” “Buried.”

By following an embedded approach, Crowdstrike Falcon enables continuous container -ai models prior to deployment, proactively discovering vulnerabilities, poisoned data sets, wrong configurations and unauthorized shadow AI.

All in all, these are factors that almost influence 64% of companies. During Runtime, Falcon uses the telemetry -driven AI of Crowdstrike, which is trained daily on trillion signals, to quickly detect and neutralize advanced threats, including fast injection, models and hidden data expansion.

Bernard clearly emphasized the unique distinguishing factor of Falcon during an interview with Venturebeat and said: “What distinguishes us is simple: we protect the entire AI living cycle. With our integration in Nvidia’s LLM NIM, we give customers the opportunity to protect models before they are implemented by the same light, running – And end points.

Bernard has further clarified and emphasized the critical runtime benefit of Falcon: “LLMS is rapidly expanding the Enterprise attack surface and the risks are really. From fast injection to API abuse we have seen how sensitive data can leak without a traditional infringement. PlatformTelemetry that set up organizations for their event.

The risk of ‘Shadow Ai’ is reminiscent

“Shadow Ai is one of the largest – and often overlooked – is running today,” Bernard warned. Shadow AI is one of the most common – and often overlooked – risks in business environments. Security teams often do not know where models are active that they build, or how they are configured – completely bypassing traditional software.

This lack of visibility creates real risk, especially in view of the sensitive data AI systems are trained on or have access to. Falcon Cloud Security discovers this hidden activity in different environments, making it visible and usable. Once you have that visibility, you can apply policy and reduce the risk. Without you flying blindly, “says Bernard.

Crowdstrike-President Michael Sentonas clearly sketched the strategic benefit clearly in an earlier Venturebeat interview, “Attackers continuously refine their techniques, and exploit the Lacunes in identity, end point and telemetry coordination. The integration of Falcon directly and reactions closes and reactions and reactions and reactions and reactions and reactions and reactions and reactions and reactions and reactions and reactions and reactions and reactions and the reactions and reactions directly and the reactions and reactions and the response and the reactions and the response to the right-time line and directly pipeline and the reactions and the reactions and the reactions and the reactions and the reactions and the reactions and the reactions to the AI ​​pipeline and right-time pipeline and the right-time pipeline and the right-time pipeline and Pijplijn and Pijplijn and pipeline pipeline and pipeline pipeline. and justifying responses and response rights and responses. ” ⁸

Following a more embedded approach to Generative AI security is a mandatory new blueprint for cisos that are confronted with the challenges of identifying and containing rapidly evolving AI threats. However, it also underlines the need for rigorous assessment: CISOs must verify whether the direct embedding in their infrastructure exactly corresponds to the individual architecture, risk exposure and strategic security objectives of their organization.

All in all, the environment of rapid adoption of AI by users and technical decision makers in workplaces looking for efficiency releases – seduced by their own personal use of consumer -oriented models such as Chatgpt, Microsoft Copilot, Anthropic Claude, Google Gemini and others – comparable to the rapid adopting risks, create a “, create” a “,”, ” -Instructions, unsecured and not -approved smartphones in the workplace during the “BYOD” era of the early 2000s and 2010.

But in this case the adoption curve of Gen AI models among users is much steeper and the technology evolves much faster, from many more players, making it even more a security mine field.

From reactive to real -time: why embedded security issues for generative AI

Traditional AI security aids that depend on external scans and interventions After deployment make companies vulnerable on the precise end points and threats when and where protection is the most critical.

Crowdstrike’s integration of Falcon Cloud Security in Nvidia’s Universal LLM NIM shifts this dynamic, which continuously encloses continuous defense in the AI ​​life cycle from development to runtime.

Bernard further explained how Falcon’s AI-SPM proactively risks reduces before implementation: “Falcon Cloud Security AI-SPM provides security and arranging IT teams earlier in the process scanning of wrong configurations, unauthorized models and policy for losing something.”

Insinning Falcon directly in the AI ​​infrastructure of NVIDIA automates compliance with emerging regulations, such as the EU AI law, as a result of which extensive model safety, traceability and auditability is an intrinsic and automated part of each implementation instead of a manual, labor-intensive task.

What Crowdstrike’s integration with nvidia means for cisos and enterprise grade gen ai security

Generative AI is rapidly expanding to Enterprise-Appropriate surfaces and the traditional security methods based on the perimeter make up.

Threats that are specific to generative models, including fast injection, data leakage and model poisoning all require deeper visibility and more precision and control. The integration of Crowdstrike with the LLM infrastructure of Nvidia is remarkable because of the architectural approach to tackling these security gaps.

For cisos, security leaders and the DevOps teams they serve, offers embedding security checks directly in the AI ​​Lifecycle offers tangible operational benefits, including the following:

• Intrinsic zero travels to scale: Automated implementation of security policy eliminates manual efforts and consistently maintains the protection of zero-trust over each AI model.
• Proactive vulnerability limitation: Identifying and neutralizing risks before Runtime considerably reduces the opportunities windows of attackers.
• Continuous runtime intelligence: Real-time telemetry-driven detection identifies and blocks threats such as fast injection, model poisoning and unauthorized data extration data.

Bernard underlined the operational need to follow a more integrative approach to generative AI security. “We are aimed at securing the models that companies themselves are building are, for example, those who are adapted to sensitive or patented data. These are not off-the-shelf risks. They require deeper visibility and stronger, made-to-measure controls around training, tuning and implementation. And customers help them to work,” he said.

Since generative AI not only becomes a distinctive factor, but a basis of business infrastructure, protection is no longer optional. Crowdstrike and the integration of Nvidia does not only add protection; It redefines how AI systems should be built to initiate the evolving Tradeecraft.