Safe, secure, assured AI

SafeAI Labs

SafeAI Labs conducts research and engineering on methods, tools, and evidence for the safe, secure, and assured use of AI-based systems—including large language models, AI agents, and physical AI—in embedded, industrial, and railway applications.

The work combines AI safety, functional safety, cybersecurity assurance, systems engineering, and standards-oriented evidence development for high-integrity cyber-physical systems.

01

Research and engineering areas

AI and LLM assurance
  • Reliability and robustness evaluation
  • Limitations and failure analysis
  • Human oversight
  • Traceable assurance evidence
AI evaluation and benchmarks
  • Benchmarks and task suites
  • Structured datasets
  • Evaluation protocols
  • Reproducible experiments
Functional safety
  • AI-assisted lifecycle support
  • Hazard and risk analysis
  • Safety requirements and functions
  • Verification and evidence production
Industrial and OT cybersecurity
  • AI-assisted threat and risk analysis
  • LLM-supported security reasoning
  • SBOM interpretation
  • IEC 62443-oriented assurance workflows
Railway safety engineering
  • Automated safety-case generation
  • Requirements and hazard analysis
  • Safety-function derivation
  • EN 50126, EN 50128, EN 50129, and EN 50716 assurance
Model-based engineering
  • UML-based system and safety models
  • Application-level safety mechanisms
  • Model generation and verification
  • Requirements-to-evidence traceability
Standards and compliance
  • IEC 61508 functional safety
  • ISO 26262 automotive safety
  • EU AI Act
  • Cyber Resilience Act

02

Selected research outputs

Topic / outputShort descriptionStatus / link
EmbSoftOTBenchBenchmarking LLMs for cybersecurity tasks in embedded and operational-technology systemsDataset on GitHub
LLMs for functional-safety analysisExamination of LLM-supported, human-in-the-loop machinery functional-safety risk analysisPublication
Functional-safety risk-assessment dataStructured dataset and methods for data-driven risk assessment in industrial functional safetyPublished research
AI-supported railway safety casesMethods to derive safety functions and develop safety-lifecycle evidence from requirements and hazard informationOngoing research
AI-generated safety mechanismsModel-based generation and verification of application-level safety mechanisms from requirements and UML inputsOngoing research
AI lifecycle and testingMethods for integrating AI-specific lifecycle phases and testing into safety-critical developmentPublication
LLM-assisted software-quality analysisScalable detection and resolution of data clumps using a modular pipeline with ChatGPTPublication

03

YearPublicationVenue
2025Clever Hans in the Loop? A Critical Examination of ChatGPT in a Human-in-the-Loop Framework for Machinery Functional Safety Risk AnalysisEng
2025On the Development and Application of a Structured Dataset for Data-Driven Risk Assessment in Industrial Functional SafetyIEEE WFCS 2025
2025Risk-Aware Prioritization of Data Clumps Refactoring in Industrial AutomationIEEE WFCS 2025
2024A Systematic Approach to Enhancing ISO 26262 With Machine Learning-Specific Life Cycle Phases and Testing MethodsIEEE Access
2025The Scalable Detection and Resolution of Data Clumps Using a Modular Pipeline with ChatGPTSoftware

All publications

04

Contact

Dr. Padma IyengharSafeAI Labs
padma@safeai-labs.com