Join the event trusted by enterprise leaders for nearly two decades. VB Transform brings together the people building real enterprise AI strategy. Learn more
Generative AI adoption has surged by 187% over the past two years. But at the same time, enterprise security investments focused specifically on AI risks have grown by only 43%, creating a significant gap in preparedness as AI attack surfaces rapidly expand.
More than 70% of enterprises experienced at least one AI-related breach in the past year alone, with generative models now the primary target, according to recent SANS Institute findings.
State-sponsored attacks on AI infrastructure have spiked a staggering 218% year-over-year, as CrowdStrike’s 2025 Global Threat Report reveals.
For CISOs, security and SOC leaders, the harsh reality is apparent. Deploying new AI models at scale exponentially expands their enterprises’ attack surfaces, and CISOs speaking on condition of anonymity have told VentureBeat traditional security tactics, strategies and technologies are challenged to keep pace. The cybersecurity industry has reached a critical inflection point: securing generative AI requires more than bolt-on tools; it demands a full architectural shift
Fortunately, CrowdStrike is also offering a new solution: On June 11 at NVIDIA’s GTC Paris event, the security firm announced that it had embedded Falcon Cloud Security directly within NVIDIA’s Universal LLM NIM (NeMo Inference Microservice). The integration secures over 100,000 enterprise-scale LLM deployments across NVIDIA’s hybrid and multi-cloud environments.
CrowdStrike’s strategic response
CrowdStrike CEO George Kurtz captured the urgency in a recent interview with VentureBeat: “Security can’t be bolted on; it has to be intrinsic. A significant part of our strategy has always been to leverage security data as a key element of our core infrastructure. You can’t secure AI without data and visibility at the deepest layers.”
“NVIDIA’s NeMo Safety provides a framework for evaluating AI risk. CrowdStrike’s threat intelligence enhances that framework by enabling security and operations teams to build guardrails around emerging AI exploit tactics – informed by what we see across trillions of daily events and real-world adversary behavior. This data advantage helps organizations assess and secure their models based on what’s actually happening in the wild,” said Daniel Bernard, Chief Business Officer, CrowdStrike, in a recent interview with VentureBeat.
Kurtz reinforced this strategic vision to Barron’s, stating clearly: “Generative AI helps us bend time. With embedded, telemetry-driven security we identify and neutralize threats at machine speed, stopping breaches probably six times faster than traditional methods.”
Bernard emphasized the significance, saying, “CrowdStrike pioneered AI-native cybersecurity, and we’re defining how AI is secured across the software development lifecycle. This latest collaboration with NVIDIA brings our leadership to the forefront of cloud-based AI, where LLMs are deployed, run, and scaled. Together, we’re giving organizations the confidence to innovate with AI, securely and at speed, from code to cloud.”
CrowdStrike embeds Falcon Security directly into NVIDIA’s AI infrastructure
By embedding Falcon Cloud Security directly into NVIDIA’s LLM NIM microservices, CrowdStrike delivers runtime protection where threats actually emerge: inside the AI pipeline itself.
“AI isn’t a standalone initiative – it’s becoming embedded across the enterprise. Unlike many cloud security vendors bolting on AI capabilities, we’ve built AI security directly into the Falcon platform. This allows us to deliver protection that’s unified across cloud, identity, and endpoint – which is critical as attackers increasingly move across domains, no longer targeting a single surface,” observes Bernard.
By taking an embedded approach, CrowdStrike is enabling Falcon to continuously scan containerized AI models prior to deployment, proactively uncovering vulnerabilities, poisoned datasets, misconfigurations, and unauthorized shadow AI.
Taken together these are factors impacting nearly 64% of enterprises. During runtime, Falcon leverages CrowdStrike’s telemetry-driven AI, which is trained daily on trillions of signals, to rapidly detect and neutralize sophisticated threats, including prompt injection, model tampering, and covert data exfiltration.
Bernard highlighted Falcon’s unique differentiator clearly during an interview with VentureBeat, saying, “What sets us apart is simple: we secure the entire AI lifecycle. With our integration into NVIDIA’s LLM NIM, we give customers the ability to protect models before they’re deployed and while they’re running—with runtime protection delivered through the same lightweight agent that already protects their cloud workloads, identities and endpoints.”
Bernard further clarified Falcon’s critical runtime advantage, emphasizing: “LLMs are rapidly expanding the enterprise attack surface, and the risks are already real. From prompt injection to API abuse, we’ve seen how sensitive data can leak without a traditional breach. Falcon Cloud Security is designed to address those gaps with real-time monitoring, threat intelligence, and platform-wide telemetry that enables organizations to stop attacks before they happen.”
The risk of ‘Shadow AI’ brings to mind the previous BYOD ‘Wild Wild West’ era of IT security
“Shadow AI is one of the biggest—and often overlooked—risks today,” Bernard warned. Shadow AI is one of the most common – and often overlooked – risks in enterprise environments. Security teams often don’t know where models are running, who’s building them, or how they’re configured – bypassing traditional software governance entirely.
That lack of visibility creates real risk, especially given the sensitive data AI systems are trained on or have access to. Falcon Cloud Security uncovers this hidden activity across environments, making it visible and actionable. Once you have that visibility, you can apply policy and reduce risk. Without it, you’re flying blind,” says Bernard.
CrowdStrike President Michael Sentonas outlined the strategic advantage clearly in a previous VentureBeat interview, “attackers continuously fine-tune their techniques, exploiting the gaps in identity, endpoint, and telemetry coordination. Falcon’s integration directly into the AI pipeline dramatically closes these gaps, giving CISOs real-time visibility and response capabilities right where attacks occur.” ⁸
Taking a more embedded approach to generative AI security represents a compelling new blueprint for CISOs who face the challenges of identifying and containing rapidly evolving AI threats. However, it also underscores the necessity for rigorous assessment: CISOs must verify whether embedding security directly into their infrastructure precisely aligns with their organization’s distinct architecture, risk exposure, and strategic security objectives.
Altogether, the environment of rapid adoption of AI by users and technical decision makers in workplaces seeking efficiency gains — enticed by their own personal usage of consumer facing models such as ChatGPT, Microsoft Copilot, Anthropic Claude, Google Gemini, and others — even without clear guidelines or permission from organizations, creates a “Wild Wild West” situation of multiple differing AI tools with differing risks, similar to the rapid adoption of unsecured and unapproved smartphones in the workplace during the “BYOD” era of the early 2000s and 2010s.
Yet in this case, the adoption curve of gen AI models among users is much steeper and the technology is evolving much faster, from many more players, making it even more of a security minefield.
From reactive to real-time: Why embedded security matters for generative AI
Traditional AI security tools that rely on external scans and post-deployment interventions leave enterprises vulnerable at the precise endpoints and threat surfaces when and where protection is most critical.
CrowdStrike’s integration of Falcon Cloud Security into NVIDIA’s Universal LLM NIM shifts this dynamic, embedding continuous defense directly into the AI lifecycle. from development to runtime.
Bernard further explained how Falcon’s AI-SPM proactively mitigates risks before deployment: “Falcon Cloud Security AI-SPM gives security and IT teams control earlier in the process—scanning for misconfigurations, unauthorized models, and policy violations before anything goes live. It helps organizations move fast without losing visibility or oversight.”
Embedding Falcon directly into NVIDIA’s AI infrastructure automates compliance with emerging regulations, such as the EU AI Act, making comprehensive model safety, traceability, and auditability an intrinsic and automated part of every deployment rather than 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 enterprise attack surfaces, straining traditional perimeter-based security methods.
Threats specific to generative models including prompt injection, data leakage, and model poisoning all require deeper visibility and greater precision and control. CrowdStrike’s integration with NVIDIA’s LLM infrastructure is noteworthy for its architectural approach to addressing these security gaps.
For CISOs, security leaders and the devops teams they serve, embedding security controls directly into the AI lifecycle offers tangible operational benefits including the following:
- Intrinsic zero-trust at scale: Automated deployment of security policies eliminates manual effort, consistently enforcing zero-trust protection across every AI model.
- Proactive vulnerability mitigation: Identifying and neutralizing risks before runtime significantly reduces attackers’ windows of opportunity.
- Continuous runtime intelligence: Real-time telemetry-driven detection rapidly identifies and blocks threats such as prompt injection, model poisoning, and unauthorized data exfiltration.
Bernard underscored the operational necessity of taking a more integrative approach to generative AI security. “We’re focused on securing the models enterprises are building themselves – especially those fine-tuned on sensitive or proprietary data. These aren’t off-the-shelf risks. They require deeper visibility and stronger, bespoke controls around training, tuning, and deployment. They require deeper visibility into prompts and responses at runtime, along with stronger, tailored controls across training, tuning, and deployment. That’s where we’re investing: securing AI with AI, and helping customers stay ahead as this technology becomes foundational to how they operate,” he said.
As generative AI becomes not just a differentiator but a foundation of enterprise infrastructure, embedded security is no longer optional. CrowdStrike and NVIDIA’s integration doesn’t just add protection; it redefines how AI systems must be built to withstand the evolving tradecraft already in motion.
[ad_2]
Source link
