Staff AI Security Engineer — Enterprise Guardrails

Affirm; Miami (FL); Remote; USD 204,000 - 264,000; Full time

Benefits offered by this job

100% health coverage

Tech stipends

Flexible time off

Stock purchase plan

Job summary

Affirm is seeking a seasoned security engineer to lead the enterprise AI security review process for AI/LLM systems, evaluating architecture, data flows and protections across tools and MCP-based systems. You will drive remediation, implement security guardrails and policy-as-code, and partner with Security, Legal, Privacy, Compliance, IT, and Engineering to enable safe AI adoption.

You will assess third-party AI capabilities, develop mitigations, contribute to incident response playbooks, and

Qualifications

• Seasoned security engineer with hands-on experience designing, evaluating, and maintaining security architecture for AI/LLM-based systems.

Responsibilities

• Lead and continuously improve the enterprise AI security review process evaluating architecture, data flows, permissions, and design of internal AI tools and MCP-based systems.
• Threat model AI/LLM-based systems and data flows for risks such as prompt injection, insecure output handling, data poisoning, and sensitive-data exposure, and drive remediation.
• Review source code, system prompts, configurations, and tool manifests; help owners build security-focused test cases and red-team scenarios.
• Design and build security guardrails and tooling for AI systems permission boundaries, authn/authz, data-handling, logging, and policy-as-code.

Job description

Affirm is seeking a seasoned security engineer to lead the enterprise AI security review process for AI/LLM systems, evaluating architecture, data flows and protections across tools and MCP-based systems. You will drive remediation, implement security guardrails and policy-as-code, and partner with Security, Legal, Privacy, Compliance, IT, and Engineering to enable safe AI adoption.

You will assess third-party AI capabilities, develop mitigations, contribute to incident response playbooks, and