Postdoctoral Associate – AI Security at University of… · Universit…

Summary: The University of Maryland's Applied Research Laboratory for Intelligence & Security (ARLIS) is seeking a Postdoctoral Associate in AI Security to conduct cutting-edge research focused on securing advanced AI systems against sophisticated adversaries. The role involves original research in adversarial machine learning, model robustness, and secure AI system design, along with collaboration with interdisciplinary teams to translate findings into actionable insights for national security. Responsibilities: - Conduct original research in AI security, including adversarial machine learning, model robustness, and secure AI system design - Develop and evaluate novel attack and defense techniques for modern AI systems, including: - Mechanistic and white-box analysis of model behavior and safety mechanisms - Multi-turn and adaptive adversarial interactions with AI systems - Security of reasoning models and agent-based architectures - Design and implement experimental frameworks for evaluating AI system vulnerabilities across deployment scenarios (e.g., open-weight, API-based, and hybrid systems) - Apply interpretability techniques (e.g., circuit analysis, feature attribution, sparse autoencoders) to understand internal model behavior and failure modes - Contribute to the development of benchmarks, evaluation methodologies, and datasets for AI security research - Collaborate with interdisciplinary teams including machine learning researchers, systems engineers, and national security domain experts - Translate research findings into actionable insights for government sponsors, including technical reports and briefings - Publish research in leading conferences and journals (e.g., NeurIPS, ICML, ICLR, IEEE S&P, CCS), consistent with program objectives Required Qualifications: - Ph.D. in Computer Science, Machine Learning, Cybersecurity, or a related technical field - Demonstrated research experience in one or more of the following areas: Machine learning (deep learning, LLMs, reinforcement learning), Adversarial machine learning or AI safety/security, Systems security, applied cryptography, or cyber operations - Strong programming skills in Python and experience with ML frameworks (e.g., PyTorch, TensorFlow) - Experience designing and executing empirical research, including experimentation and evaluation - Ability to work in a collaborative, interdisciplinary research environment - Ability to obtain and maintain a U.S. security clearance Preferred Qualifications: - Familiarity with white-box threat models and evaluation of open-weight AI systems - Experience with MLOps or large-scale training infrastructure, including distributed training, GPU clusters, or ML experimentation platforms - Knowledge of AI system deployment architectures, including RAG systems, multi-agent systems, or tool-augmented models - Experience with adversarial evaluation frameworks, red-teaming methodologies, or benchmark development - Experience with mechanistic interpretability and/or alternative approaches to understanding model internals (e.g., activation analysis, circuit-level reasoning, representation learning) - Background in national security applications, including work with DoD, IC, or federally funded research programs - Record of publications in top-tier conferences or journals Required Skills: Machine learning, Adversarial machine learning, AI security, Cybersecurity, Systems security, Applied cryptography, Python, PyTorch, TensorFlow, Empirical research, MLOps, Distributed training, GPU clusters, Red-teaming methodologies, Mechanistic interpretability, Activation analysis, Circuit-level reasoning, Representation learning