CrowdStrike is actively recruiting AI researchers to develop security agents that leverage test-time compute, a technique where AI models spend more computational resources during inference to reason through complex problems rather than relying solely on pre-training. The cybersecurity giant processes nearly three trillion events per day and is now building the next generation of agentic systems to automate analyst workflows and improve threat detection reliability.

What Is Test-Time Compute and Why Does It Matter for Security?

Test-time compute, also called inference-time scaling, represents a fundamental shift in how AI systems approach problem-solving. Rather than making decisions instantly, these models can allocate additional computational resources during inference to search through possibilities, verify answers, or reason through multiple approaches before settling on a final response. For cybersecurity, this means AI agents can spend more time analyzing suspicious activity, cross-referencing threat patterns, and validating their conclusions before alerting analysts.

The job posting explicitly lists familiarity with "inference-time scaling and test-time compute, including search, self-consistency, verifier-guided decoding, and long chain-of-thought" as a preferred qualification. This signals that CrowdStrike views these techniques as essential to building AI systems that security teams can actually trust with high-stakes decisions.

How Is CrowdStrike Structuring Its AI Research Around Agentic Systems?

CrowdStrike's Data Science team is expanding to focus on building AI agents that combine multiple reasoning loops, tool calling, and memory management to automate security analyst procedures. The company is seeking researchers with deep expertise in several interconnected areas:

• Post-training and reinforcement learning: Fine-tuning large language models (LLMs) using supervised fine-tuning and reinforcement learning techniques like RLHF (reinforcement learning from human feedback), RLAIF (reinforcement learning from AI feedback), PPO (proximal policy optimization), GRPO, and DPO (direct preference optimization) to improve model reliability on real security tasks
• Agent architecture and planning: Designing systems that combine planning and reasoning loops, tool and function calling, and retrieval and memory management to create increasingly complex workflows that mimic how human analysts work
• Rigorous evaluation frameworks: Establishing objective criteria for benchmarking agentic systems through evals, LLM-as-judge pipelines, and trajectory-level metrics with statistical rigor to ensure agents perform reliably in production
• Inference optimization: Optimizing prompts and inference to maximize model performance, including the application of test-time compute techniques to get the most value from each inference call

The emphasis on "establishing objective criteria for benchmarking agentic systems" underscores a critical challenge in the field: measuring whether AI agents actually work better when given more reasoning time, and at what computational cost.

What Technical Skills Are Required to Build These Systems?

CrowdStrike is seeking researchers with PhD-level depth in modern machine learning, including mastery of LLM training fundamentals such as architecture, optimization, tokenization, data, and scaling behavior. The ideal candidate will have hands-on experience with the full machine learning stack, including GPUs, PyTorch, and common LLM training and serving tools like Hugging Face Transformers, DeepSpeed, FSDP, vLLM, TGI, and SGLang.

Beyond core machine learning skills, the role requires expertise in building training data and environments, including synthetic data generation, agent trajectories and rollouts, and task simulators. Researchers must also understand agent safety and guardrails, including sandboxing, jailbreak resistance, and reliability for autonomous systems that operate with minimal human oversight.

How Does This Hiring Signal Broader Trends in AI-Powered Security?

CrowdStrike's focus on test-time compute and agentic systems reflects a broader industry recognition that simply scaling model size is insufficient for complex, high-stakes domains like cybersecurity. By investing in researchers who understand how to make AI systems reason more carefully during inference, the company is positioning itself to build security agents that can handle ambiguous threat scenarios, verify their own conclusions, and provide analysts with transparent reasoning traces.

The job posting also emphasizes interpretability and failure analysis, noting that researchers should have "a knack for interpretability and failure analysis, diagnosing why a model or agent fails, not just that it does." This reflects a maturation in how security-focused AI teams think about deployment: understanding not just whether a system works, but why it fails and how to improve it.

CrowdStrike's emphasis on combining machine learning expertise with cybersecurity domain knowledge suggests that the future of AI-powered security lies not in generic AI systems, but in specialized agents trained on real threat data and tuned to the specific workflows of security operations teams. By hiring researchers who can bridge both worlds, the company is signaling its commitment to building AI systems that security professionals will actually use and trust.