AI Security & Governance Engineer

About Astound Digital

At Astound Digital, we are pioneers in the digital landscape, dedicated to transforming how the world interacts with technology, data, and creativity. Our role as trusted advisors in the digital landscape empowers the world’s most innovative brands with frictionless, end-to-end customer experiences. We are known for our comprehensive solutions, proven expertise, and collaborative, nimble approach that instills confidence in our clients. We embrace the power of Artificial Intelligence (AI) to drive innovation, efficiency, and better outcomes for our clients and employees. Join us to navigate and lead in the ever-changing digital world, where your impact will extend beyond the ordinary.

Job Purpose

At Astound Digital, we have identified, piloted, and implemented an internal AI platform that uses AI tooling across our client delivery to enable our teams to collaborate, develop, and manage, and drive project and internal work more effectively and efficiently . Over the past eighteen months, this work has produced a unified LLM access layer for our engineering organization with tiering and spend limits, a knowledge and tooling platform on GCP built around Vertex AI Search and MCP, and a growing set of client commitments regarding how data is handled inside AI systems. Most client delivery ships on Salesforce, with our Anthropic practice leveraging Claude on top of it. Currently, all of this relies on architectural decisions and manual oversight.

Astound Digital is looking for someone to own security and AI engineering end-to-end: turn existing practice into a managed function, address the technical risks in the AI platform, and position our AI security as a core asset in enterprise client conversations rather than a source of unexpected questions.

This is a greenfield role with no legacy constraints or ready-made processes. This is an opportunity to own, define, and refine security and AI engineering across the processes, platforms, and firm.

Key Responsibilities

AI platform security

Threat modelling for RAG and agentic scenarios: indirect prompt injection through indexed content, trust boundaries between data sources, context isolation between client projects
Reviewing and controlling MCP integration permissions: which tools agents can reach, who approves extending that perimeter, how changes get re-reviewed
Designing and implementing guardrails and data-leakage controls across AI pipelines
Security review of AI functionality before it ships to production
Working with the platform team on logging and monitoring for AI services: what must be logged, which detections to build, what alerting looks like
Reviewing connected app configuration and OAuth scopes for AI integrations into Salesforce orgs, ours and our clients'
Client-facing AI delivery

Security review of the AI solutions we deliver to clients, on Salesforce (Agentforce, Data Cloud, MCP connectors into client orgs) and on the Anthropic platform (Claude API, Claude Enterprise, Claude Code, Claudeforce plugins), against an agreed turnaround so review is a gate and not a queue

Owning the standard answers on how client data is handled across the platforms we deliver on: retention, residency, model routing, what the provider sees and what it keeps, so the practices and sales are not reconstructing them per deal
Maintaining the inventory of client-facing AI deployments with the model provider, routing path and billing path recorded for each
Scoping and managing the per-engagement red teaming we buy in, and folding findings back into the guardrail baseline

AI governance and risk

Owning internal AI policy: model access tiers, acceptable use, the exceptions process, while coordinating with the respective organizations such as Legal and HR
Defining the sanctioned path to each model provider, including direct API access where a partner program or client contract requires it, and keeping routing choices visible to the practices that depend on them
Maintaining the inventory of AI systems and vendors, and reassessing risk on a regular cadence