AI Data-Lineage Regulatory Evidence Counsel — Capital Markets

The mandate

The risk committee’s standing question is whether the group can reconstruct a consequential AI-assisted decision as it occurred, not as today’s data catalogue suggests it should have occurred. Trading supervision, communications review and client-risk models use features assembled through rapidly changing pipelines. Management reports lineage coverage, yet exceptions, point-in-time joins and analyst interventions remain difficult to explain under supervisory questioning.

The adviser will provide two days each month for committee preparation, evidence challenge and senior-management clinics, and join five scheduled Board Risk Committee sessions. A request arising from a material model incident or regulatory examination receives an initial written view within one US business day. Formal testimony, investigative work or model validation requires separate board authority and cannot be absorbed into cadence.

The appointment begins in January 2027 and runs for ten months. At month eight, the chair will assess whether directors and accountable executives can apply the agreed evidence standard without continuing support. One extension of up to four months may be approved by the full committee; unused days expire monthly and do not lengthen the term automatically.

The adviser has no line authority, executive responsibility, model approval, compliance delegation or audit role. Advice tests sufficiency and clarity but cannot certify regulatory compliance, direct remediation or speak for the company. Model owners, Compliance and the CRO retain their distinct decisions, and the minutes must record where a management risk acceptance differs from the adviser’s view.

Up to two other appointments are allowed, subject to disclosure of work for broker-dealers, exchanges, model vendors, data providers, regulators or enforcement counsel. An active engagement involving the same supervisory matter, dataset or disputed transaction creates a conflict unlikely to be cured by recusal. Investments and expert-network work touching shortlisted suppliers must also be declared before papers are issued.

Why the board wants this voice

The committee understands model governance and regulatory accountability but lacks a practitioner who has reconstructed feature evidence across fast-changing capital-markets systems. Management’s completeness measures do not answer the point-in-time questions directors expect from supervisors. The chair wants a seasoned challenger who can distinguish a lineage graph from legally and operationally credible decision evidence.

What you will own

• Press the committee to define which AI-assisted decisions require point-in-time reconstruction and what evidential confidence is proportionate to consequence.
• Test whether lineage covers source entitlements, effective-time joins, feature code, model release, human intervention, output use and subsequent correction.
• Challenge completeness percentages that omit manual extracts, ephemeral streams, vendor transformations, fallback logic or records retained outside the central catalogue.
• Shape escalation thresholds for broken lineage, unverifiable historic decisions and pipeline changes made during regulatory or litigation preservation periods.
• Probe whether data owners, model owners and business users understand their separate responsibilities when evidence spans organisational boundaries.
• Frame questions for supervisory readiness exercises that expose retrieval time, interpretive ambiguity and inconsistent narratives before an external request arrives.
• Coach directors to separate technical provenance, record retention, model validation and legal sufficiency rather than treating one artefact as all four.

Candidate qualifications

• Held senior data, model-risk, compliance-technology or regulatory evidence responsibility within a broker-dealer, exchange or complex financial institution.
• Reconstructed point-in-time AI or quantitative decision evidence for a supervisor, enforcement matter, audit or consequential internal investigation.
• Understands event-time and processing-time differences, changing reference data, feature stores, manual overrides and retention across heterogeneous trading systems.
• Challenged a superficially complete lineage programme whose measures failed to support an actual historical decision or supervisory question.
• Advised boards while maintaining clear separation among management accountability, independent validation, compliance judgement and legal privilege.
• Managed conflicts involving financial institutions, regulators, data vendors, technology providers and counsel in a portfolio advisory practice.

Non-negotiables

• Can attend the five New York committee meetings in person despite the otherwise remote work mode and respond within the agreed incident window.
• Will disclose all regulated-entity work, enforcement support, expert testimony, investments and relevant vendor relationships before appointment.
• Accepts that the role cannot validate models, certify compliance, direct employees or assume an executive’s regulatory accountability.
• Must demonstrate point-in-time evidential reconstruction in capital markets; generic data-governance credentials do not meet the threshold.

1. 49 words maximum. Describe a point-in-time model decision you reconstructed and the lineage element a central catalogue failed to retain.
2. 49 words maximum. Which current regulated-entity, vendor, legal or supervisory engagements would require disclosure to this board?
3. 49 words maximum. How would you test whether lineage completeness supports a real regulatory question rather than an architecture diagram?