Senior Product Security Engineer

Responsibilities:

Band A:

• Build and run the security agents. Design, build and operate the security agents on the Claude Agent SDK with MCP tool integrations. Starting with PR Sentinel and DepGuard, then outward across the lifecycle.
• Wrap the scanners we buy so they become one coherent, low-noise signal rather than seven disconnected dashboards.
• Evaluate and red-team your own agents. False negatives and prompt injection against a security agent are your problem to catch.
• Automate the first response to customer security questionnaires, so an enterprise buyer gets a substantive answer in hours.
• Claude Agent SDK

Band B:

• Secure what we build. Threat-model high-risk features, new data flows, AI capabilities, auth and identity changes before code exists.
• Own SAST, secret scanning, dependency and supply-chain security in CI, with SLAs on critical and high findings.
• Run vulnerability assessment in-house: DAST, API security testing, and hands-on penetration testing of our own products.
• Review application code and API designs, and make cross-tenant isolation structurally impossible to get wrong.
• Scope and manage external specialists where they earn their cost-crown-jewel penetration tests, independent assessments and hold their output to a standard.
• Raise the level of the engineers around you: security champions in every squad, practical training, and reference material people actually use.
• Supply the technical evidence behind our ISMS and customer audits. The certification program is owned by our security team in Japan; you make it defensible.

Band C:

• Secure what we ship (AI / LLM). Red-team Mico's own AI agents: prompt injection, jailbreaks, tool abuse, data exfiltration through model output.
• Build guardrails and regression gates so a prompt or model change cannot silently weaken defences.
• Work to OWASP LLM Top 10andMITRE ATLAS, and turn the results into evidence our enterprise customers can read.

Band D:

• Secure what we hold (Data and PII). Classify data, then enforce it: masking and redaction in logs, non-production and analytics and before anything reaches a model.
• Implement just-in-time, least-privilege access across personal data, with audit trails that stand up to scrutiny.
• Build tokenisation and a mediated data-access path so no service or person reads sensitive stores directly.

You will not be doing this alone. How the work is actually carried out.

• Agents carry the volume. This is the entire point of our agentic approach to security. One engineer cannot review every pull request, triage every advisory and re-test every AI feature by hand, so you build the agents that do the first pass, and you spend your time on the judgment calls. Leverage is the design here, not an afterthought.
• Squads own their own fixes. You are not the person patching a dozen codebases. Security champions are named in every squad; they are the first readers of agent output and the people who land the remediation. You build the system that makes them effective.
• Work directly with VPs. Our VP and security specialists in Japan will support you on roadmap, strategy and development.
• The work is sequenced, not simultaneous. Look at the first six months below: the early work is coverage and classification. AI red-teaming and data-access controls come afterwards, once the foundation carries its own weight. Nobody is expected to start four programs in week one.
• The function grows with you. This is an opportunity to be a first member of the security team and to grow that team with you. Being first here means shaping a team and the opportunity to become its lead in time.

And what this role is not:

Here is what sits outside this role, and where it actually lives.

• You will not run corporate IT. Devices, office network, helpdesk and identity administration belong elsewhere.
• You will not own ISMS / ISO 27001 audit and evidence management. Our security team in Japan owns the certification program.
• You support it technically; you are not the auditor.
• You will not be the compliance function. We will set this up separately, together.

What you will own:

Your first ninety days: the engineering pipeline:

• You take ownership of scanning coverage across our repositories: static analysis, dependency and supply-chain checks, and secret detection. You own data classification across our primary stores, and the threat-modelling gate that high-risk features pass through. You name the security champions in each squad and set how they work with you.

Months four to six: agents in the engineers' path:

• You own the review agent that gives every pull request an explanation and a suggested fix, and the orchestration layer that turns many scanners into one prioritised queue. You run assessment in-house DAST, API security testing and penetration testing alongside the external specialists we engage.

Requirements:

• 5-8 years in security, the majority of it in application or product security for a software product, not solely network or IT security.
• You write production code. Python, Go or TypeScript, and comfortable in Shell. You have automated security work rather than only operating tools that someone else built. (non-negotiable)
• Hands-on across the application security core: secure code review, threat modelling, API and web vulnerability assessment, OWASP Top 10 and API Top 10 and CVE/OSV-driven dependency and supply-chain security.
• Cloud security on AWS in practice: IAM, network boundaries, and real use of GuardDuty, Security Hub or CloudTrail.
• Security wired into CI/CDSAST, SCA and secret scanning in pipelines you owned, including the unglamorous work of getting developers to accept them.
• Daily practical use of AI coding agents such as Claude Code in your own work.
• Working fluency in English. Our engineering organisation spans more than twenty nationalities across India and Japan.

What would help:

• AI / LLM security: prompt injection, jailbreak and tool-abuse testing; OWASP LLM Top 10 MITRE ATLAS; harnesses such as Promptfoo, Garak or PyRIT. Rare, and genuinely valued. If you don't have it yet, this is the role where you build it.
• PII engineering at scale: classification, masking, tokenisation, data-access proxies; exposure to APPI or GDPR.
• Security due diligence for regulated buyers: questionnaires, customer audits, trust centres.
• Certifications such as OSCP, GWAPT, CISSP. Useful signal, never a substitute for the work.
• Japanese language ability. Welcome, not required.