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

Situation

An examiner request for the current model inventory put hiring-tool adverse-impact tables in front of HR analytics governance lead in a retailer using generative AI in customer service. This AI Governance / Bias and Training Data close is a generative-AI incident is from hiring-tool adverse-impact tables, and the live options are A generative-AI incident is a policy breach, A model defect.

Decision

HR analytics governance lead in a retailer using generative AI in customer service must choose A generative-AI incident is a policy breach / A model defect using hiring-tool adverse-impact tables after an examiner request for the current model inventory.

Hypotheses to test

• HR analytics governance lead can defend A generative-AI incident is a policy breach from hiring-tool adverse-impact tables after an examiner request for the current model inventory in a AI Governance challenge.
• HR analytics governance lead cannot defend A generative-AI incident is a policy breach from hiring-tool adverse-impact tables; A model defect is what the extract actually supports after an examiner request for the current model inventory.
• An examiner request for the current model inventory never reached the population in hiring-tool adverse-impact tables — reopen intake, do not close a generative-AI incident is.
• Two facts in hiring-tool adverse-impact tables after an examiner request for the current model inventory conflict for HR analytics governance lead; hold this Bias and Training Data file.

Analysis required

• Walk the model input/output path recorded in hiring-tool adverse-impact tables and mark each hop approved, shadow, or unlogged.
• Verify data provenance and the human-oversight gate HR analytics governance lead can actually point to.
• Walk the model input/output path recorded in hiring-tool adverse-impact tables and mark each hop approved, shadow, or unlogged.
• For this AI Governance Bias and Training Data file, read hiring-tool adverse-impact tables against an examiner request for the current model inventory and write the one fact that would move a generative-AI incident is for HR analytics governance lead.

Recommendation

Choose A generative-AI incident is a policy breach / A model defect on this AI Governance / Bias and Training Data packet (hiring-tool adverse-impact tables after an examiner request for the current model inventory). The follow-on Bias and Training Data action is what HR analytics governance lead does next: implement the option, assign an owner, and log the missing fact.

Command returns

• Bottom-line AI Governance option on a generative-AI incident is, then the evidence in hiring-tool adverse-impact tables, then the action for HR analytics governance lead
• Hypothesis scorecard against hiring-tool adverse-impact tables: supported / rejected / untestable
• Regulatory or exam hook Bias and Training Data would cite
• Bias and Training Data finding in hiring-tool adverse-impact tables that a second reviewer can re-perform