AI is changing the speed and scale of cyber defense, and the speed and scale of the adversary.
As AI becomes embedded across government, critical infrastructure, and enterprise environments, defenders need the ability to inspect, test, adapt, and secure the systems they depend on. This requires leveraging the best AI capabilities, including open models, open harnesses, frontier models, and shared tooling that allow the security community to evaluate how AI systems behave, understand how adversaries may use them, and collectively improve defenses.
CrowdStrike is joining NVIDIA and other industry leaders as an inaugural partner in the Open Secure AI Alliance, a coalition focused on advancing AI safety and security through open innovation, shared research, and collective defense. Through this work, CrowdStrike will share insights on accelerating AI defense with open models, such as those outlined in the CrowdStrike 2026 Global Threat Report.
Trustworthy AI Starts with Security
AI safety is not solved by evaluating a model in isolation. In cybersecurity, defenders need to understand how highly capable models behave when they are used for real security work, and how those same capabilities can be weaponized by adversaries. Open models, a valuable defense tool, are today being tuned, evaluated, and run in production to actively improve the security of enterprises.
That is where the harness becomes critical. The harness determines what context an AI system can access, how it reasons through a task, what actions it is allowed to take, the tooling used, and how its conclusions are validated. It is the difference between raw model output and security intelligence an analyst can trust.
CrowdStrike has seen this directly. In vulnerability research testing, using the same class of frontier AI models with different harnesses produced dramatically different results. A generic approach yielded false-positive rates approaching 80%. Applying a CrowdStrike-developed security harness reduced that rate to approximately 20% while maintaining strong vulnerability discovery capability. The takeaway is clear: the model matters, but the way it is operationalized determines whether it improves defense or creates noise.
That is why open models, open harnesses, and shared tools are important. They give defenders a way to test AI systems against real adversary behaviors, identify failure modes, validate safeguards, and improve techniques across the broader security community. This is how AI safety becomes measurable through evidence, repeatability, and continuous improvement.
The Open Secure AI Alliance brings that work into the open, where defenders can build on shared research instead of duplicating effort in silos. To support open development through initiatives like the Open Secure AI Alliance, CrowdStrike is developing techniques to use open models as detectors for AI and agentic attacks that safeguard enterprises and their environments. Join us at Fal.Con to learn more.
The Right Model for the Right Mission
The most effective AI strategy is not defined by one model or deployment approach. It is defined by matching the right capability to the right security mission.
CrowdStrike works with the world’s leading frontier AI providers. Advanced reasoning models and frontier foundation models are critical to pushing the boundaries of what AI can do for cyber defense. Recent CrowdStrike work with frontier models, including Anthropic’s Mythos Preview and OpenAI TAC, demonstrated what is possible when advanced AI is paired with the right security harness by surfacing more than 2,400 vulnerability findings across 43 products in a matter of days.
At the same time, building AI security systems that operate at global scale requires open models that can be fine-tuned to specific security use cases, deployed efficiently, governed appropriately, and operated without unnecessary exposure of sensitive telemetry. It also requires economics that make enterprise-grade AI security practical for the organizations and agencies that need it most.
CrowdStrike’s collaboration with NVIDIA on Nemotron open models demonstrates the production value of open models. After fine-tuning Llama Nemotron Super 49B for natural-language-to-CQL translation, CrowdStrike achieved 96% valid query accuracy, outperforming tested closed-source frontier alternatives on this specialized security task. Today, open models give defenders deeper control over how AI is adapted, evaluated, and deployed across real SOC workflows and support the safety of critical production environments. They make AI operationally viable across the security workflows where speed, scale, cost, and control matter.
Why This Matters for Government and Critical Infrastructure
Government agencies and critical infrastructure operators are both major consumers of AI and central to shaping how AI is developed, governed, and secured. As these organizations adopt AI for mission-critical use cases, they need transparent systems that can be inspected, evaluated, adapted, and trusted.
Open models and open tooling support those priorities by enabling independent evaluation, and allowing agencies and operators to tune AI to mission-specific requirements while maintaining control of sensitive operational data. They also give researchers, vendors, academic institutions, and allied governments a broader foundation for identifying vulnerabilities, developing mitigations, and sharing intelligence.
That distributed model of defense has long been one of cybersecurity’s greatest strengths. The AI era makes it even more important.
Defense in the AI Era
AI safety is not achieved through opacity. It requires rigorous testing, secure infrastructure, strong governance, human oversight, and broad technical scrutiny. Open models, open harnesses, and shared tools provide the foundation for that work.
The Open Secure AI Alliance is building that foundation by driving open research, practical tooling, and collective expertise to help make AI more secure and effective for defenders.
CrowdStrike is proud to join NVIDIA and other industry leaders as an inaugural partner. The AI era demands a new kind of defense, and building it will take the full strength of the defender community.