The TÜV AUSTRIA Trusted AI framework provides an end‑to‑end audit and certification methodology for machine learning systems in safety‑critical environments. Developed since 2019 in collaboration with the Institute for Machine Learning at Johannes Kepler University Linz, the Software Competence Center Hagenberg, and the joint venture TRUSTIFAI, it translates the obligations of the European Union Artificial Intelligence Act into concrete, testable criteria.
Core Concept: Functional Trustworthiness
The framework defines a statistically grounded application domain and establishes risk‑based minimum performance requirements. Independent statistical testing on newly sampled data provides transparent and reproducible evidence of real‑world model quality.
Three Pillars
- Secure Software Development – ensures robust engineering practices
- Functional Requirements – aligns assessment with the full lifecycle of an artificial intelligence system
- Ethics and Data Privacy – incorporates lawful, fair, and transparent processing
Practical Insights
Real‑world audits highlight common pitfalls: data leakage, poorly defined domains, overlooked biases, and missing controls for distribution drift.
Outlook
The framework integrates robustness, algorithmic fairness, and post‑certification obligations. By aligning technical best practices with emerging European standards, it offers regulators, providers, and users a practical path to legally compliant and certifiable artificial intelligence systems.
