KRAI Index — Responsible AI Evaluation
The KalpaneAI Responsible AI Index (KRAI) evaluates large language models against Dutch and EU expectations across legal compliance, privacy, robustness, fairness, transparency, and human oversight.
Leader Tier
Illustrative examples of models that may score strongly on KRAI dimensions.
*Ratings shown below are illustrative examples for demonstration purposes only and do not represent official or validated benchmark results.
Apertus-8B
Strong Responsible AI posture across legal, privacy, and transparency dimensions.
TildeOpen
Good alignment with governance expectations, with relatively balanced model behaviour.
Llama-3
Competitive capability, though deployment posture depends heavily on controls and use context.
Mistral
Promising enterprise use, but requires structured governance and closer evaluation per deployment.
What KRAI Measures
- Legal compliance with EU AI Act and AP guidance
- Privacy and data protection alignment with GDPR
- Accuracy, reliability, and hallucination control
- Bias, non-discrimination, and fairness testing
- Transparency, disclosure, and AI system limitations
Why It Matters
- Helps organisations compare LLMs beyond raw capability
- Supports safer AI adoption under Dutch and EU standards
- Creates a practical governance view for deployment decisions
- Links technical evaluation with regulatory accountability
- Enables clearer Responsible AI positioning for enterprise use
Foundation models, enterprise AI tools, open-weight ecosystems, vendor transparency, and deployment readiness.
GDPR principles, lawful processing, access control, traceability, and AP-aligned governance.
Reliability, validation layers, guardrails, fallback mechanisms, and control of plausible but incorrect outputs.
Fairness testing, transparency, non-discrimination, and mitigation of skewed or harmful outputs.
Meaningful oversight, accountability, interpretability, escalation paths, and trusted decision support.