Interim Chief AI Officer — Model Governance Recovery

The mandate

An automated eligibility model produced materially different outcomes for protected customer cohorts after a feature pipeline change, and the AI head was dismissed for bypassing release review. The model is paused, but forty-six other production models lack a consistent inventory, evaluation record or accountable business owner.

The interim must start within three weeks and serve a fixed ten months. Recruitment for a permanent Chief AI Officer begins after the portfolio risk classification is approved, with the final six weeks reserved for successor induction.

Handover is complete when the affected model has been repaired or retired with customer consequences resolved, every production model has an accountable owner and risk tier, high-risk releases pass independent evaluation, and the successor has chaired one deployment committee. The control must work without the interim's manual intervention.

The interim may suspend models, mandate evaluation, choose technical remediation and direct ₹11 crore from the approved AI assurance budget. New high-impact use cases, training-data purchases above ₹4 crore, permanent leadership hires, customer remediation above ₹5 crore and risk acceptance for severe bias need committee or board approval.

General data-platform replacement, basic analytics and commercial ownership of AI products are outside scope. The role governs models that influence consequential decisions but does not take over Legal's interpretation of discrimination or Product's revenue targets.

Why this seat is open

The bypass made prior self-attestation by model teams untenable. Internal candidates either built the affected system or own product targets dependent on other unreviewed models. The board wants a temporary executive who can establish credible controls before selecting a long-term AI leader.

What you will own

• Bound the affected population and decide whether the model is repaired, constrained or permanently withdrawn.
• Create an enterprise model inventory with use, owner, training lineage, risk tier, evaluation and live-monitoring status.
• Set deployment gates for performance, subgroup outcomes, robustness, explainability, privacy and human override.
• Approve independent evaluations and reject tests whose data or operating conditions do not represent production use.
• Establish live drift and harm triggers with named suspension authority and customer-remediation routing.
• Certify every high-risk model against the new standard or provide the committee with a dated retirement plan.
• Transfer the model register, accepted risks, evaluation library, talent assessment and next review calendar to the successor.

Candidate qualifications

• Held Chief AI Officer, responsible-AI head or enterprise machine-learning director authority over production models.
• Contained and remediated a consequential model failure involving bias, drift, safety or data lineage.
• Built model-risk tiers and deployment controls across multiple product teams and use cases.
• Directed data scientists, ML engineers, evaluators and governance specialists at enterprise scale.
• Can translate statistical evidence into board decisions about suspension, remediation and customer consequence.
• Has operated with Indian privacy and discrimination considerations plus global enterprise AI assurance expectations.

Non-negotiables

• Available in Bengaluru within three weeks.
• No current financial interest in the external model-evaluation or foundation-model vendors.
• Will disclose material model incidents occurring under prior authority.
• Must preserve independent evaluation and documented human override for high-risk decisions.

1. 49 words maximum. Confirm your start date and any model-vendor conflict.
2. 49 words maximum. Describe a production model you suspended and the subgroup or safety evidence that triggered action.
3. 49 words maximum. Which evaluation must pass before a repaired eligibility model returns to production?