AI risk lies in the use, not the model

A model name does not determine risk. Purpose, role, affected people, decision impact, data flow, autonomy and consequence do.

A bright central AI model cube connects to four distinct deployment contexts: internal desk, chat interface, decision gate and safety-critical control; only the consequential path carries an amber warning marker.

01 · SYSTEMIs the real process described with data, tools, people and downstream decisions?

02 · CONTEXTAre purpose, role, affected people and actual decision influence supported by evidence?

03 · DUTYWere the AI Act, data protection and sector rules assessed separately and against the correct dates?

CORE METHOD

SYSTEM → CONTEXT → DUTY → CONTROL

Capture the deployed system, classify its context, justify applicable duties and translate them into verifiable controls.

THE PURPOSE-CHANGE TEST

Use the same model in two cases: internal drafting and applicant prioritisation. Compare affected people, decision impact, data, roles, duties and stop criteria.