This list contains only the countries for which job offers have been published in the selected language (e.g., in the French version, only job offers written in French are displayed, and in the English version, only those in English).
Our client is a leading global investment management company headquartered in London. It manages over $228 billion in assets and serves institutional investors, pension funds, wealth managers, and other sophisticated clients worldwide. The firm specializes in quantitative investing, alternative investments, systematic trading strategies, and technology-driven asset management. Data science, machine learning, and AI are core components of its investment and research processes. As part of our collaboration we will focus on two foundational capabilities required to enable safe and scalable AI adoption across the enterprise: Agentic Security and AI-Ready Data Foundations. What project we have for you We define how autonomous agents authenticate, obtain scoped access, and operate safely across a large, regulated financial estate where the runtime security model genuinely does not exist yet. The value, and the danger, of agentic AI is set by what an agent can reach: an agent that inherits a full user context and long-lived secrets has an effectively unlimited blast radius. Your job is to close that gap. This is a hands-on senior-level role for a security engineer who remains deeply technical and actively ships production code, operating at the intersection of enterprise IAM, platform engineering, and the emerging domain of agentic AI security. You will design and build the IaC-driven, self-service identity patterns, credential flows, and onboarding standards that make the secure way the easy way, across high-velocity teams that have long governed themselves.
Job Responsibility
Design and ship IaC-driven, self-service identity patterns that roll out firm-wide without requiring a full Active Directory cleanup first
Define the currently undefined agentic runtime security model: containerised code execution, permission delegation to agents, and MCP-based tool access
Lead the transition from long-lived secrets toward ephemeral, time-based, risk-scored credentials, scoped to task duration and issued via JWT / OIDC
Layer LLM / software guardrails (policy-as-text plus human review) on top of whatever hard guardrails are feasible across the estate
Establish an opinionated onboarding standard (e.g. mandatory MCP interfaces) and win adoption through better defaults and developer experience, not mandate alone
Design SIEM integration, behavioural baselining, and anomaly detection for agentic workflows, and centralise siloed audit logs to satisfy both security and regulatory requirements
Take bounded beachheads (for example, authenticate users and then delegate scoped access to internal systems) from vague to delivered
Requirements
6+ years in security architecture and/or platform engineering, with a track record of shipping production code. Principal / Staff-level depth, ideally in a high-velocity or quant / financial-services engineering culture
Deep, mechanical command of modern identity and authorisation: OIDC / OAuth2 / JWT — token issuance flows, claims design, and delegation / impersonation patterns
Hands-on HashiCorp Vault experience, including dynamic / short-lived secrets and the realities of migrating off long-lived tokens without breaking a large application estate at once
Keycloak policy modelling, ideally with the Terraform-driven configuration the firm already uses
Strong Terraform / IaC fluency — enough to design repeatable, self-service patterns that others adopt, rather than bespoke per-team setups
Working knowledge of the Active Directory + Entra legacy reality: nested groups, LDAP-backed role mapping, and the distribution-list-as-permission-group failure mode — able to design around the mess pragmatically
Nice to have
Real exposure to agentic / LLM systems and why they change the threat model — an agent actively probes and exploits standing permissions rather than stumbling onto them. That removes the “security through obscurity” cushion humans relied on
Familiarity with MCP as an integration / onboarding standard, and at least one agent harness (Claude Agent SDK preferred)
Experience with just-in-time, task-scoped delegation versus standing access, and risk-gated credential issuance (e.g. a short-lived token issued against a CrowdStrike-style risk score)
Behavioural baselining / anomaly detection for workloads — defining “normal” for a recurring workflow and catching deviation at volume
SIEM integration and action attribution: distinguishing an agent’s action from the human whose credentials it borrowed
Financial-services audit literacy
Consulting or client-facing / pre-sales experience