# EU AI Act Risk Classifier
Fine-tuned DistilBERT model that classifies AI system descriptions into EU AI Act risk tiers.
## Intended Use
Classifies a natural-language description of an AI system into one of five categories:
- `minimal_risk`
- `limited_risk`
- `high_risk`
- `prohibited_risk`
- `review` (flagged for human review)
Intended as a first-pass triage tool to support (not replace) human compliance review under the EU AI Act.
**Out of scope:** This model does not provide legal advice and should not be used as the sole basis for a compliance decision.
## Training Data
- Dataset size: 1,120 labeled examples
- Labels sourced via a dual-labeling pipeline (LLM-assisted first pass + deterministic rule-tree cross-check) with human review on flagged conflicts
## Training Procedure
- Base model: DistilBERT
- Learning rate: 3e-5
- Epochs: 15 (best checkpoint at epoch 13, selected by validation accuracy)
- Batch size: 8
- Max sequence length: 256
## Limitations
- The "review" …
