Lead Research Scientist - AI for Cyber-Physical Risk

Established nearly two centuries ago, FM is a leading mutual insurance company whose capital, scientific research capability and engineering expertise are solely dedicated to property risk management and the resilience of its policyholder-owners. These owners, who share the belief that the majority of property loss is preventable, represent many of the world’s largest organizations, including one of every four Fortune 500 companies. They work with FM to better understand the hazards that can impact their business continuity to make cost-effective risk management decisions, combining property loss prevention with insurance protection.

FM is hiring a Lead Research Scientist to build the team's AI and machine learning capability across Cyber Security, Robotics and Digital Twins research. The role answers one question for FM: how do AI-enabled industrial systems create new paths to physical damage and business interruption, and how can AI help us measure that risk?

The role has two sides. First, study AI as a source of risk: models embedded in OT/ICS, autonomous robots and digital twins that can fail, drift or be attacked. Second, use AI as a research tool: foundation-level modelling, data pipelines and analytics that turn lab and loss data into risk insight for underwriting and engineering.

Across all three domains

• Set and own the AI research roadmap for the Automation and Intelligent Systems research group.

• Build shared AI foundations: data pipelines, feature stores, model training and evaluation environments, and MLOps practices that can be reused by all three domains.

• Develop an AI risk taxonomy for industrial settings: failure modes, adversarial threats, model drift, over-reliance and loss of human oversight, each mapped to physical damage and business interruption outcomes.

• Design experiments in the lab that make AI failure consequences measurable, not just theoretical.

• Turn findings into underwriting guidance, risk engineering recommendations, loss-prevention data sheets and publications.

• Track relevant standards and regulation (EU AI Act, ISO/IEC 42001, ISO/IEC 23894, NIST AI RMF, IEC 62443, ISO 10218) and assess what they mean for insured risk.

• Mentor researchers and engineers in the team and raise overall AI literacy.

Cyber security (OT/ICS)

• Research adversarial ML against AI used in industrial control: evasion, data poisoning, model extraction and supply-chain compromise of models.

• Assess AI-based anomaly and intrusion detection for OT networks, including false-negative risk and attacker evasion.

• Study how LLM agents and AI copilots in engineering and operations tools widen the attack surface (prompt injection, tool misuse, unsafe autonomous actions).

• Model cyber-to-kinetic scenarios and aggregation exposure where one AI model or vendor is deployed across many insured sites.

Robotics and autonomous systems (Physical AI)

• Conduct cutting-edge applied research on AI on robotic systems — AGVs, AMRs, quadruped and humanoid robots — including embodied AI, 3D mapping and localization (e.g. SLAM, S-Graphs), machine learning, and autonomy for robotic technologies operating in industrial environments.

• Advance image processing, computer vision, and sensor data fusion techniques, linking them to potential failures while characterizing the associated hazards.

• Assess vision-language-action and other foundation models for robotics, and the risk introduce.

• Support robotic experimental approaches, testing and safety work with AI-specific test methods.

Digital twins

• Build physics-informed and hybrid ML models (surrogates, reduced-order models) of industrial processes and equipment.

• Use digital twins to simulate AI failures and safely, and to generate synthetic data for rare loss events.

• Assess the fidelity, validation and trustworthiness of twins that operators use for decisions, including twin–asset desynchronisation risk.

• Explore twins as risk engineering tools: what-if analysis, scenario generation and loss estimation.

Machine learning and AI foundations

• Strong command of supervised, unsupervised, self-supervised and reinforcement learning.

• Deep learning architectures: CNNs, RNNs/LSTMs, transformers, graph neural networks and generative models.

• Time-series modelling and anomaly detection on sensor, process and network data.

• Working knowledge of foundation models and LLMs: fine-tuning, retrieval-augmented generation, evaluation and agentic systems.

• Probabilistic modelling, Bayesian methods and uncertainty quantification.

AI robustness, safety and security

• Adversarial machine learning: attack and defense methods, robustness testing, red-teaming of models.

• Model evaluation beyond accuracy: calibration, out-of-distribution detection, drift, failure analysis.

• Explainability and interpretability methods (for example SHAP, attribution methods, mechanistic analysis).

Engineering and data

• Expert Python; fluent in PyTorch and/or JAX, plus scikit-learn and the scientific Python stack.

• MLOps: experiment tracking, reproducible pipelines, model versioning, containers and GPU compute.

• Building and curating datasets from messy industrial sources; labelling strategy and synthetic data.

Qualifications

EDUCATION AND EXPERIENCE

Education

• PhD in Computer Science, Machine Learning, Electrical/Electronic Engineering, Robotics, Applied Mathematics or a closely related field.

• A Master's degree with an exceptional research and publication record may be considered in place of a PhD.

Experience

• 5+ years of post-doctoral or industry research experience in machine learning or AI (8+ years if no PhD).

• Demonstrated work applying AI to cyber-physical, industrial, safety-critical or autonomous systems.

• Publication record in recognised venues (for example NeurIPS, ICML, ICLR, IEEE S&P, USENIX Security, ICRA, IROS, or relevant IEEE/ACM journals).

• Experience leading research projects end to end, from proposal to delivered results.

• Experience working with industrial partners or turning research into applied outcomes.

Languages

• Fluent English, written and spoken. French, German or Luxembourgish is an advantage.
  

DESIRED SKILLS

Domain exposure (at least one of the three)

• Robotics: ROS 2, perception stacks, SLAM, motion planning, sim-to-real transfer, functional safety (ISO 10218, ISO 13849, ISO 3691-4).

• Digital twins and simulation: physics-informed neural networks, reduced-order modelling, co-simulation, tools such as NVIDIA Isaac/Omniverse, Gazebo, Simulink or Modelica.

• OT/ICS environments: PLCs, SCADA, DCS, industrial protocols (Modbus, OPC UA, PROFINET), IEC 62443.

Advanced AI topics

• Safety of LLM agents and autonomous decision systems, including prompt injection and tool-use risk.

• Causal inference and causal discovery for root-cause and loss analysis.

• Federated or privacy-preserving learning for data from multiple client sites.

• Edge AI: model compression, quantization and deployment on embedded hardware.

• Formal verification or runtime assurance of neural networks.

Risk and industry context

• Quantitative risk methods: fault trees, FMEA, Bayesian networks, Monte Carlo simulation, extreme-value analysis.

• Familiarity with AI governance frameworks (EU AI Act, ISO/IEC 42001, NIST AI RMF, MITRE ATLAS, OWASP Top 10 for LLMs).

• Understanding of insurance, loss prevention or reliability engineering.

• Experience with public research funding, industry–academic partnerships or PhD co-supervision.

PERSONAL AND SOFT SKILLS

• Systems thinking: connects a models to physical work outcomes and to the loss that follows.

• Clear communication: explains AI risk to all stakeholder levels without jargon, in writing and in person.

• Scientific rigour and healthy scepticism: challenges AI hype and states the limits of evidence plainly.

• Collaboration across disciplines: works well with cyber specialists, roboticists, simulation engineers and insurance professionals.

• Self-direction: shapes open problems into research questions and delivers without close supervision.

• Mentoring and knowledge sharing: builds AI skills in colleagues.

• Pragmatism: balances research depth with results the business can use on a realistic timeline.

• Integrity and discretion: handles confidential client data and sensitive vulnerability findings responsibly.

• Curiosity: keeps up with a fast-moving field and brings new methods into the team.

Less qualified candidates may be considered for a Sr Research Engineer Or Scientist position.