The opportunity
Replit is the agentic software creation platform that enables anyone to build applications using natural language. With millions of users worldwide, Replit is democratizing software development by removing traditional barriers to application creation.
What you'll do
AI Agent Security Strategy & Technical Execution
AI Agent Security Blueprint: Define the long-term vision and architectural patterns for securing autonomous agent workflows, Model Context Protocol (MCP) integrations, multi-turn reasoning loops, and multi-agent coordination.
Runtime Guardrails & Policy Enforcement: Architect and deploy dynamic input/output guardrail systems, semantic firewalls, and real-time intent verification filters to prevent goal hijacking, system prompt leaks, and indirect prompt injections.
Agent Execution & Tool Sandboxing: Partner with Infrastructure and AppSec teams to design secure, short-lived, micro-isolated environments (e.g., microVMs, WebAssembly, container sandboxes) where agents can dynamically execute code, run shell commands, and interact with host operating systems safely.
Agentic Threat Modeling & Red Teaming: Conduct specialized threat modeling against non-deterministic systems. Lead automated and manual AI red-teaming initiatives to uncover vulnerabilities in RAG context pipelines, vector stores, and tool-calling interfaces.
Identity, Delegation & Least Privilege: Architect fine-grained permission models and step-up Human-in-the-Loop (HITL) authorization patterns for agents acting on behalf of users—ensuring agents never exceed the delegated privileges of the end user or misuse API keys.
What they're looking for
- Context & Memory Boundary Security: Design strict data isolation and poisoning prevention controls for vector databases, retrieval-augmented generation (RAG) pipelines, and long-term conversation memories across multi-tenant workloads.
- Risk Management & Cross-Functional Enablement
- Maintain the AI Security Source of Truth: Define, document, and maintain authoritative secure design patterns and SDKs for engineering teams building internal or customer-facing AI agent capabilities.
- Contribution to Risk Register: Identify, quantify, and document non-deterministic and agentic security risks (e.g., OWASP Top 10 for LLMs & AI Agents, MITRE ATLAS), ensuring they are accurately represented in the Cybersecurity Risk Register.
