Vice President, Agentic AI Risk Engineering

The mandate

A financial technology platform is introducing agents that can assemble information, invoke tools and propose actions across servicing workflows. The central problem is not whether the models can generate plausible answers, but whether their permitted autonomy remains safe when tools, context and business conditions change. This specialist Vice President establishes the controls through which agent capabilities are bounded, tested and reviewed.

The role is open-ended permanent employment, with an eighteen-month initial agenda covering autonomy classification, intervention evidence and controlled release governance. Its specialist executive perimeter combines AI risk leadership with cross-product programme execution. Product engineering retains software delivery and infrastructure ownership; risk owners retain policy decisions, while this function connects those responsibilities to observable agent behaviour.

The leader will distinguish a recommendation from an action, and a constrained tool call from an open-ended sequence whose consequences are difficult to predict. Permission design, human checkpoints and escalation conditions must be linked to actual workflow risk. A generic statement that a human remains involved is inadequate unless the operating evidence shows what the person sees, when intervention occurs and whether it can still prevent the relevant harm.

Within the approved governance framework, the Vice President sets assurance test priorities, control evidence standards and release recommendations for the AI risk function. Product deployment decisions and policy exceptions follow the designated approval route. The remit excludes autonomous credit sanctioning policy, enterprise cybersecurity ownership and certification that every model output is correct; uncertainty must be managed through defined operating boundaries rather than erased in a control paper.

During the first year, each material agent workflow should have an approved autonomy perimeter, reproducible intervention tests and an accountable owner for changes to its tools or instructions. The following six months strengthen monitoring, incident learning and control maintenance as products evolve. Ongoing leadership then preserves the connection between declared safeguards and demonstrated behaviour, rather than treating initial launch approval as permanent assurance.

What you will own

• Establish an autonomy classification register that records permitted actions, tool access and consequence severity for each workflow, enabling reviewers to distinguish constrained assistance from delegated execution.
• Decide the risk assurance tests required before release recommendation, including tool misuse, missing context and intervention failure scenarios that reflect the actual financial servicing process.
• Develop a human checkpoint evidence standard showing what reviewers can observe and stop, challenging controls that rely on nominal approval after the relevant action has already occurred.
• Govern change reviews for agent instructions, connected tools and permission sets, ensuring modifications that alter practical autonomy return through the appropriate risk and product approval route.
• Build a control observation pack linking test results, operating incidents and unresolved limitations, making it possible for senior governance to see where declared safeguards lack demonstrated evidence.
• Lead cross-functional incident learning reviews for the AI risk perimeter, translating observed failure patterns into specific control changes without claiming authority over independent policy or software implementation owners.
• Develop assurance specialists through workflow case studies and evidence reviews, creating leadership depth that can challenge emerging agent behaviours rather than depend on one expert's interpretation of general frameworks.

Candidate qualifications

• Demonstrate substantial financial-services AI governance, innovation programme or risk-control leadership, with practical work on autonomy frameworks or generative AI guardrails. Explain a workflow where action sequencing created a risk not visible in a single model response, and describe the permission, intervention or evidence change you personally shaped to address it.
• Bring the ability to translate recognised AI risk frameworks into controls that can be observed and tested. Relevant experience includes classifying use cases, documenting limitations and coordinating engineering and compliance evidence. You must distinguish an internal control design from a legal interpretation of a jurisdiction's AI obligations, obtaining qualified specialist review when the question requires it.
• Show rigorous assurance judgement under incomplete model predictability. Describe how you designed adverse tests, assessed human intervention effectiveness and recorded residual risk without promising universal correctness. The role does not demand ownership of every infrastructure component, but it does require informed challenge of the tool permissions and operating interfaces that determine practical agent autonomy.
• Establish senior programme and team leadership across product, risk and engineering stakeholders. Provide evidence of a contested release recommendation, a clearly documented approval boundary and follow-through after launch. Experience should show constructive restraint when specialist AI expertise did not confer authority over credit policy, cybersecurity or wider enterprise technology decisions.

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 9 October 2026. Mandate reference CVU-PER-2026-IND-194.