AI Assurance Evidence Architecture Principal

The mandate

An AI enablement organisation has credible control intentions but review evidence is scattered across product demonstrations, risk papers and test records with inconsistent ownership. The project will design an AI Assurance Evidence Architecture for three agreed financial-services workflow families. It will show how a claim is supported, challenged and maintained, rather than provide blanket certification of model behaviour or legal compliance.

The engagement begins on 26 October 2026, spans five months and reserves four days weekly. Named artifacts include a claim-to-evidence map, a test specification library and a reviewer operating pack. The architecture must distinguish observed test results, management representations and unresolved limitations so governance users can understand the strength of a control conclusion without reconstructing all the underlying product work.

Milestone one on 7 December 2026 establishes the evidence inventory and claim taxonomy. Milestone two on 8 February 2027 supplies the mapped assurance architecture and sampled test specifications for the agreed workflows. The final milestone on 26 March 2027 delivers completed reviewer replay results, evidence maintenance rules and the accepted transfer pack covering the whole five-month term.

The AI Governance Head and Risk Assurance Director accept outputs through traceability and reproducibility tests. Every sampled material claim must lead to an identifiable test or limitation, a named evidence owner and a review trigger; an internal reviewer must rebuild the sample conclusion from the supplied pack. Acceptance does not depend on favourable model outputs, a regulator's response or resolution of legal questions outside the design scope.

The sponsor provides product documentation, existing safeguards, test artifacts and four representatives available for workshops and replay. The three accepted stages release 20%, 40% and 40% of the fee respectively. Software implementation, independent penetration testing and legal opinions are excluded. Adding workflow families or requiring new technical testing beyond the specification library requires written change control and an agreed additional evidence plan.

What you will own

• Inventory the evidence supporting the three workflow families, distinguishing test artifacts, governance decisions and management assertions so the initial architecture does not treat unlike evidence as interchangeable proof.
• Develop a claim taxonomy that links autonomy, intervention and lifecycle safeguards to specific assurance questions, making the review perimeter understandable to product owners and independent risk specialists.
• Design the claim-to-evidence map with ownership and change triggers, identifying unsupported claims explicitly rather than constructing a reassuring control narrative around incomplete technical records.
• Specify reproducible tests for selected material safeguards, explaining expected observations and limitations so implementation teams understand what a successful result would and would not establish.
• Build a reviewer operating pack that permits reconstruction of sampled conclusions, preserving the source versions and decision context needed to interpret evidence after the original product workshop has ended.
• Facilitate internal reviewer replay across two representative adverse cases, repairing missing lineage or ambiguous instructions before the joint sponsors assess the final evidence architecture.
• Deliver maintenance and scope-change rules that identify when new tools, altered use or changed control claims require evidence reconsideration, keeping future ownership with the sponsor's designated operating functions.

Candidate qualifications

• Demonstrate practical AI governance or financial-services assurance design experience, including control claims, test evidence and review traceability. Describe a claim that appeared well supported until you examined the underlying artifact, and explain how you changed the conclusion or test specification without claiming broader assurance than the available evidence permitted.
• Bring familiarity with generative AI guardrails, agent autonomy concepts and recognised risk management approaches sufficient to design meaningful evidence questions. Relevant proof should show translation from high-level expectations to observable tests. The appointment does not require unrelated infrastructure ownership, but it does require informed judgement about what technical specialists must demonstrate for a claim to be credible.
• Show independent facilitation across product and risk teams, with precise written scope and limitations. Explain how you resolved disagreement about acceptable evidence, who owned the final decision and how you preserved an unresolved issue in the architecture. Legal interpretation and regulated policy opinions must remain with qualified owners rather than be implied by the assurance terminology used.
• Establish four-day weekly capacity for five months and a record of transferable specialist artifacts. Provide an example of reviewer replay that failed because lineage or instructions were incomplete, and describe the repair. Consulting success here depends on reproducible acceptance and internal maintenance capability, not favourable test results or continuing dependence on your judgement.

Application

Applications for this mandate are received in one way only: through the India Board Terminal's application process. It is automated end to end. Your Executive Passport travels to the mandate holder in its confidential form, your answers to the three questions below are read before anything else in your file, and every stage that follows is recorded on your applications page.

There is no address to write to and no intermediary to call. The mandate holder reads what the Terminal delivers and nothing else, which is what keeps the process the same for every applicant and keeps your name out of it until you release it. Applications close on 13 October 2026. Mandate reference CVU-CON-2026-IND-198.