Generative AI adoption has seen a significant increase of 177% in the past two years, according to a report by Gartner. However, enterprise security investments focused on AI risks have only grown by 43%, creating a notable gap in preparedness as AI attack surfaces continue to expand rapidly.
Recent findings from the SANS Institute indicate that more than 70% of enterprises have experienced at least one AI-related breach in the last year, with generative models being the primary target. Additionally, state-sponsored attacks on AI infrastructure have surged by 218% year-over-year, as revealed in CrowdStrike’s 2025 Global Threat Report.
The escalating threats posed by AI vulnerabilities have left CISOs and security leaders facing a challenging reality. Deploying new AI models at scale significantly increases the attack surfaces of enterprises, outpacing the capabilities of traditional security tactics and technologies. As a result, securing generative AI now requires a comprehensive architectural shift rather than just relying on bolt-on tools.
In response to these evolving challenges, CrowdStrike has introduced a new solution. At NVIDIA’s GTC Paris event on June 11, the security firm announced the integration of Falcon Cloud Security into NVIDIA’s universal LLM NIM, securing over 100,000 enterprise-scale LLM deployments across hybrid and multi-cloud environments.
CrowdStrike’s CEO, George Kurtz, emphasized the importance of intrinsic security, stating that security cannot be added as an afterthought but must be integrated at the core. The collaboration with NVIDIA aims to leverage threat intelligence to enhance AI risk evaluation and enable organizations to build guardrails around emerging AI exploit tactics based on real-world adversary behavior.
By embedding Falcon Cloud Security directly into NVIDIA’s LLM NIM microservices, CrowdStrike provides runtime protection within the AI pipeline itself. This approach allows for continuous scanning of containerized AI models before deployment to proactively identify vulnerabilities, misconfigurations, and unauthorized shadow AI activities.
The integration of Falcon into NVIDIA’s AI infrastructure enables real-time monitoring, threat intelligence, and platform-wide telemetry to detect and neutralize sophisticated threats during runtime. This strategic response not only ensures the security of AI models but also automates compliance with emerging regulations such as the EU AI Act.
For CISOs and security leaders, embedding security controls directly into the AI lifecycle offers operational benefits such as intrinsic zero-trust at scale, proactive vulnerability mitigation, and continuous runtime intelligence. As generative AI becomes foundational to enterprise infrastructure, embedded security is no longer optional but essential for safeguarding against evolving threats.
In conclusion, the collaboration between CrowdStrike and NVIDIA redefines how AI systems should be built to withstand the growing challenges of AI security risks. By securing AI with AI and providing tailored controls across the development and deployment process, organizations can stay ahead in protecting their AI assets from malicious actors.
