Auditability & Traceability
Auditability ensures that an AI system can be analysed, verified and reproduced. It is a central requirement of the AI Act and a foundational pillar of Responsible AI.
Why Auditability Is Essential
An AI system must allow its functioning, data, parameters and decisions to be traced. Without auditability, compliance, human oversight and trust become impossible.
The AI Act imposes strict requirements for documentation, logs and reproducibility.
Traceability of Data and Models
Mandatory Logs
High‑risk systems must maintain detailed logs: data used, model versions, parameters, metrics and decisions.
Model Versioning
Every version of a model must be identifiable and reproducible. A decision function must be re‑executable with the same parameters and the same data.
If parameters or data are not preserved, auditability is lost.
Reproducibility
Reproducibility ensures that an AI system produces the same results when re‑executed under the same conditions. It is a central requirement of the AI Act: a model must be verifiable, auditable and reliably re‑executable.
Operational Definition
A system is reproducible if, for the same dataset, the same pipeline and the same configuration, it generates exactly the same outputs before and after re‑execution, redeployment or audit.
What This Concretely Implies
- input data must be versioned and accessible;
- model parameters must be preserved and documented;
- the training and inference pipeline must be stable and controlled;
- the execution environment (libraries, GPU, seeds, dependencies) must be fixed or reproducible;
- sources of non‑determinism must be identified and controlled.
Why It Is Difficult
Modern ML pipelines contain many non‑deterministic elements: random initializations, parallelization, GPU execution, library versions, implicit preprocessing. Without strict control, two “identical” runs may produce different results.
AI Act Requirements
- Complete and timestamped logs
- Detailed technical documentation
- Verifiable human oversight
- Risk management
- Demonstrable reproducibility
- Independent auditability
These requirements are at the core of the MathIAs+ offerings: Software, Academy and Governance.
Building a Responsible AI Culture
Responsible AI is not only a regulatory requirement — it is a strategic capability. MathIAs+™ Academy helps your teams master modern, sovereign practices.
Explore the Academy