Security Engineer
- Location
- San Francisco, CA
- Employment
- Full Time
- Work model
- On-Site
- Level
- Mid
- H-1B history
- 1 approvals (FY2023)
- Posted
Skills
About this role
Anrok is the leading tax automation platform enabling businesses to expand globally without compliance complexity. As the digital economy has grown 6x over the last decade, software businesses have gone from not worrying about sales tax to needing to monitor exposure, calculate rates, and file returns across 20+ US states and many countries worldwide. This creates a critical bottleneck for companies that should be able to transact with customers everywhere. Anrok eliminates this complexity by connecting with billing and payment systems to automate tax monitoring, calculations, and filing end-to-end. Our unified platform handles the ever-changing maze of tax laws at the municipal, state, and federal levels, so companies can focus on growth, not compliance. Our customers include: 40% of Forbes Top 50 AI companies 20% of Forbes Top 100 Cloud companies Top companies like Notion, Anthropic, and Cursor We're making compliant digital commerce a reality for companies big and small, backed by over $100M from leading investors including Spark Capital, Sequoia, Index, and Khosla Ventures.
The role
Most security teams grow by adding people: another analyst for alerts, another engineer for design reviews, another person for triage. We’re taking a different approach at Anrok. As our first security engineer, you’ll build a strong product and cloud security program, covering secure development, a hardened cloud, detection and response, and incident leadership. AI agents handle the repeatable work so you can focus on the judgment calls. What you build in your first year will shape how far a small team can go as Anrok grows. Anrok sits in the path of our customers’ revenue and holds sensitive financial, tax and identity data. For them, security is a crucial part of the product they’re buying. What AI-native means here Repeat work goes to agents. If you've done a task twice, the third time should be a workflow. Agents propose, people decide. Anything that changes production, closes a finding or reaches a customer needs a human approver and leaves an audit trail. Agents get the same scrutiny as engineers. Each one has its own identity, least-privilege access, logged actions and evals before we rely on it. We measure leverage. The question is how much coverage one engineer gives a growing company, not how many tickets get closed. What you'll build Product and application security A secure development process engineers trust: scanning in CI, AI-assisted review on pull requests that touch payments, credentials or auth, and threat models for our highest-risk designs. Findings triage that mostly runs itself, plus our pentest and vulnerability disclosure programs, with each fix agreed with the owning team. Securing AI at Anrok Security for the AI features we ship, including our MCP and agent integrations: tool permissions, prompt injection, data boundaries between customers, and testing before every launch. Guardrails for how Anrok uses AI internally, covering which tools can reach which data, coding agents, AI vendors, and internal tools built by non-engineers. Cloud and infrastructure A hardened cloud baseline written as code and checked continuously, with secrets that rotate automatically. Tight access to customer data across production and internal tools, with an owner and scoped permissions for every service account, API key and agent identity. Detection and response Detections across cloud, identity and source control, with an AI triage layer that investigates routine alerts and escalates the ones that matter. Our incident response plan, and leading security incidents from triage through postmortem. Across the team Technical controls for SOC 2 and ISO 27001, built with our GRC Lead so they produce their own evidence. Support for enterprise security reviews, and helping shape the Security team’s roadmap and hiring. Your first year One year in, success looks like: Every critical and high finding has an owner and is fixed within SLA, even