COMPEL Glossary / risk-based-classification
Risk-Based Classification
Risk-based classification is an approach to AI governance that applies different levels of regulatory requirements, oversight, and governance controls based on the potential risk of harm from the AI application.
What this means in practice
The EU AI Act establishes four tiers: prohibited practices (social scoring, subliminal manipulation), high-risk systems (employment, credit, healthcare, law enforcement -- subject to extensive requirements), limited-risk systems (chatbots requiring transparency disclosure), and minimal-risk systems (most AI applications with few requirements). COMPEL's own Agent Governance layer uses a complementary four-tier classification: low risk, medium risk, high risk, and critical risk, combined with the six-level autonomy spectrum to produce a composite governance intensity score. Risk-based classification ensures that governance effort is proportional to potential harm -- preventing both under-governance of dangerous systems and over-governance of benign ones.
Why it matters
Risk-based classification ensures governance effort is proportional to potential harm, preventing both under-governance of dangerous systems and over-governance of benign ones. As the EU AI Act and other regulations adopt tiered approaches, organizations need their own classification systems that align with regulatory tiers while addressing internal governance needs. Without proportional classification, governance becomes either dangerously lax or prohibitively burdensome.
How COMPEL uses it
COMPEL's Agent Governance layer uses a four-tier classification (low, medium, high, critical risk) combined with the six-level autonomy spectrum to produce a composite governance intensity score during the Model stage. This classification determines the rigor of testing, monitoring, and human oversight required. The Evaluate stage verifies classification accuracy, and the Governance pillar maps COMPEL tiers to EU AI Act tiers for regulatory alignment.
Related Terms
Other glossary terms mentioned in this entry's definition and context.