AI Red Team Specialist (Generative AI & Agentic Systems)

BLEND Localization; Indonesia; On-site; IDR 150,000,000 - 300,000,000; Full time

Job summary

BLEND Localization is assembling a bench of offensive security specialists for an AI red-team programme. You will attack generative AI systems, including models, LLM apps, and agents that hold tools and act on prompts.

We seek real exposure to generative AI security, with a focus on producing actionable findings and client-ready reports. Fluency in English and verifiable security credentials are required.

Qualifications

• Hands-on experience with offensive security, especially AI systems and agents.
• Ability to assess prompts, guardrails, and data exfiltration vectors in RAG contexts.
• Proven ability to document findings with steps, root causes, severity, and remediation.

Responsibilities

• Probe LLM applications and agentic systems for exploitable behaviour: prompt injection, guardrail bypass, system prompt extraction, tool abuse, data exfiltration.
• Demonstrate impact with actionable jailbreaks and reliable proof of concept.
• Draft client-ready findings: reproduction steps, root cause, severity, and OWASP Top 10 mapping for LLM apps.
• Operate within defined scope and document results clearly.

Skills

Generative AI security

Familiar with vulnerability research

English writing proficiency

Education

Degree in CS or Information Security

Tools

Python

NodeJS

Job description

BLEND is building a bench of offensive security specialists for an AI red-team programme with one of our clients. The target is not a web application. It is a generative AI system: models, LLM applications, and agents that hold tools and act on what they are told.

You will attack those systems and write up what you find. If you have spent years breaking web applications and have started turning that instinct on LLMs, this is the work.

What you will do

• Probe LLM applications and agentic systems for exploitable behaviour: prompt injection, both direct and indirect, guardrail and policy bypass, system prompt extraction, tool and function-call abuse, data exfiltration through RAG pipelines, privilege escalation across agent boundaries.
• Prove impact, not just output. A jailbreak that reaches nothing is a curiosity.
• Write findings the client can act on: reproduction steps, root cause in the agent or tool design, severity, and a mapping to the OWASP Top 10 for LLM Applications.
• Work to a defined scope and stay inside it.

What we are looking for

• Real exposure to generative AI security, professional or self-directed. Personal research counts here, and we will ask you to show it.
• Fluent written English. The report is the deliverable.
• Evidence we can verify: a CVE credit, a public bug bounty profile, a vendor acknowledgement, a published write-up or advisory, a conference talk, or an issuer verification link for the certifications you hold.
• Willingness to complete an identity check and sign an NDA before touching any client system.

Nice to have

• Around 2 years of hands-on web penetration testing. If your track record is shorter but backed by published AI security work, apply anyway and tell us what you have done.
• Offensive security certifications in the OSCP / OSWE / OSCE3 or GIAC families.
• Python or NodeJS, enough to build your own harness.
• A Hack The Box, TryHackMe or CTFtime profile.
• A degree in computer science or information security.