Microsoft Principal Applied Scientist - Security AI Models 

Taiwan, Taoyuan City 

101354806

The Security Models Training team builds and operates the large-scale AI training and adaptation engines that power Microsoft Security products, turning cutting-edge research into dependable, production-ready capabilities. As a, you will lead end-to-end model development for security scenarios, including privacy-aware data curation, continual pretraining, task-focused fine-tuning, reinforcement learning, and rigorous evaluation. You will drive training efficiency on distributed GPU systems, deepen model reasoning and tool-use skills, and embed responsible AI and compliance into every stage of the workflow. The role is hands-on and impact-focused, partnering closely with engineering and product to translate innovations into shipped experiences, designing objective benchmarks and quality gates, and mentoring scientists and engineers to scale results across globally distributed teams. You will combine strong coding and experimentation with a systems mindset to accelerate iteration cycles, improve throughput and reliability, and help shape the next generation of secure, trustworthy AI for our customers.

Required Qualifications:

• Bachelor's Degree in Statistics, Econometrics, Computer Science, Electrical or Computer Engineering, or related field AND 8+ years related experience (e.g., statistics, predictive analytics, research)
  • OR Master's Degree in Statistics, Econometrics, Computer Science, Electrical or Computer Engineering, or related field AND 6+ years related experience (e.g., statistics, predictive analytics, research)
  • OR Doctorate in Statistics, Econometrics, Computer Science, Electrical or Computer Engineering, or related field AND 5+ years related experience (e.g., statistics, predictive analytics, research)
  • OR equivalent experience.
• 3+ years experience presenting at conferences or other events in the outside research/industry community as an invited speaker.
• Proven track record training and shipping large language models or multimodal models for production scenarios, including continual pretraining and task specific adaptation.
• Proficiency in Python and PyTorch, with hands-on experience building and debugging large-scale training jobs.

Ability to meet Microsoft, customer and/or government security screening requirements are required for this role. These requirements include, but are not limited to the following specialized security screenings:

Microsoft Cloud Background Check:
- This position will be required to pass the Microsoft background and Microsoft Cloud background check upon hire/transfer and every two years thereafter.

Preferred Qualifications:

• Security domain experience in one or more areas: security operations, threat intelligence, malware analysis, vulnerability and posture management, anomaly detection, phishing and fraud detection, or cloud identity and access.
• Experience with distributed training and scaling techniques, for example DeepSpeed, FSDP, ZeRO, model and pipeline parallelism, mixed precision, and profiling.
• Experience with privacy preserving ML including differential privacy concepts, privacy risk assessment, and utility measurement on privatized data.

Certain roles may be eligible for benefits and other compensation. Find additional benefits and pay information here:Microsoft will accept applications for the role until October 21, 2025.

• Lead the full modeling lifecycle for security scenarios, from data ingestion and curation to training, evaluation, deployment, and monitoring.
• Architect and optimize large-scale distributed training on GPU clusters, improving throughput, reliability, and cost efficiency.
• Design and implement privacy-preserving data workflows, including anonymization, templating, synthetic augmentation, and quantitative utility measurement.
• Develop and maintain fine-tuning and adaptation recipes for transformer models, including parameter-efficient methods and reinforcement learning from human or synthetic feedback.
• Establish objective benchmarks, metrics, and automated gates for accuracy, robustness, safety, and performance, enabling repeatable model shipping.
• Collaborate with engineering and product teams to productionize models, harden pipelines, and meet service-level objectives for latency, throughput, and availability.
• Mentor applied scientists and engineers, promote strong documentation and experiment hygiene, and foster a culture of rapid iteration grounded in responsible AI principles.
• Contribute to thought leadership by staying current with the latest AI advances and translating promising techniques into practical, measurable impact.