Platform Security [US]

About the role

OUR MISSION

Rebuild how the world works, to make institutions work better for the people they serve.

ABOUT BRAIN CO.

Brain Co. builds AI-native operating systems for large, regulated institutions. Each system is built for a specific industry, powered by agents that push real workflows forward. Underneath it all is Atlas, our proprietary platform that keeps customers in control, secure by design, and never locked into one model.

WHY NOW

Brain Co. is entering its next phase of production deployments on a national scale with an elite team built from Palantir, Google, Meta, and Nvidia, and a growing footprint across government, insurance, health, and financial services.

Joining now means shaping both the company and a new category of applied AI. Every project here ships to production and is expected to create measurable customer value and impact.

You'll work alongside exceptional peers on some of the hardest problems in applied AI. It’s the kind of work you'll still be proud of in ten years from now.

About the Role:

We’re looking for a Platform Security Engineer to build the guardrails that keep our AI agents safe to run on real customer data. This isn’t a policy or compliance role—it’s a builder role. You’ll design and ship the code that constrains what an agent can access, do, and expose: scoped credentials, data access boundaries, action validation, and audit trails, so product teams can put agents in front of sensitive government, healthcare, and enterprise data with confidence.

You’ll work as a software engineer embedded with our product and agent-platform teams—writing production code, not just policy—to make the secure path the only path an agent can take.

What You’ll Build & Ship

- Design and build the guardrail services that mediate actions an AI agent takes, scoped permissions, tool-call validation, and hard limits on what an agent can read, write, or send

- Write production code for data access controls that keep customer PII and sensitive records inside approved boundaries, even when an agent is orchestrating the request

- Build reusable guardrail libraries and SDKs so product engineers can drop data protection and permissioning into new agent workflows without reinventing it each time

- Design detection and containment for agent-specific failure modes, prompt injection, tool misuse, data exfiltration attempts, and build automated tests and red-team harnesses to catch them before production

- Instrument agents with tamper-evident audit logs and decision trails so every customer-data access is explainable after the fact

- Partner with product, platform and ML engineering to review new agent capabilities before launch and flag where guardrails are missing

- Own the developer experience for guardrails: clear APIs, documentation, and low-friction integration so engineers adopt controls instead of routing around them