Assess whether a generative-AI incident is a policy breach or a model defect

Situation

EU AI Act implementation manager owns a generative-AI incident is inside an insurer scoring claims with a third-party model with shadow-IT chatbot connected to customer PII as the only packet. A customer complaint that a chatbot invented a refund policy is what changed the clock for this AI Governance Policy and Oversight file.

Decision

EU AI Act implementation manager in an insurer scoring claims with a third-party model must choose A generative-AI incident is a policy breach / A model defect using shadow-IT chatbot connected to customer PII after a customer complaint that a chatbot invented a refund policy.

Hypotheses to test

• The population in shadow-IT chatbot connected to customer PII is the one a customer complaint that a chatbot invented a refund policy named, so A generative-AI incident is a policy breach follows for this Policy and Oversight file.
• The population in shadow-IT chatbot connected to customer PII is adjacent only to a customer complaint that a chatbot invented a refund policy; A model defect is the honest AI Governance call.
• An insurer scoring claims with a third-party model already contained a customer complaint that a chatbot invented a refund policy before shadow-IT chatbot connected to customer PII arrived; no new Policy and Oversight path.
• Provenance on shadow-IT chatbot connected to customer PII after a customer complaint that a chatbot invented a refund policy is broken; do not pick A generative-AI incident is a policy breach or A model defect yet.

Analysis required

• Split policy-or-governance failure from a model defect using prompts, outputs, and human edits in shadow-IT chatbot connected to customer PII.
• Reproduce the incident row in shadow-IT chatbot connected to customer PII and say whether it ever touched production data.
• Split policy-or-governance failure from a model defect using prompts, outputs, and human edits in shadow-IT chatbot connected to customer PII.
• For this AI Governance Policy and Oversight file, read shadow-IT chatbot connected to customer PII against a customer complaint that a chatbot invented a refund policy and write the one fact that would move a generative-AI incident is for EU AI Act implementation manager.

Recommendation