Volunteer Team: App Development Team

Time Commitment: Volunteer

Volunteer Availability: Virtual

AI safety lead volunteer opportunity at iLevelUP. Fully remote, unpaid work for eligible applicants in the United States. This opening is for the AI Evaluation, Red Team & Safety Lead role.

AI safety lead: remote volunteer opportunity

Make trustworthy AI behavior something the team can test, explain, and improve.

Fully remote | Unpaid volunteer | United States

About iLevelUP

iLevelUP is a free, AI-powered, game-based learning experience being developed by Believe in Me to help young people explore education and career possibilities and build self-confidence. The product is in limited beta. The learner is the main character, and iLevelUP is the guide.

AI safety lead role overview

The official role is AI Evaluation, Red Team & Safety Lead. iLevelUP seeks a remote volunteer AI safety lead to test AI experiences before young people use them. You will evaluate the full interaction, including prompts, retrieved information, tools, voice, character behavior, and the way a learner experiences the result.

This is a hands-on role in a limited-beta product. You will turn broad concerns into reproducible tests, prioritized fixes, and clear recommendations that engineers and leadership can act on.

What you will own

You own the AI evaluation program and the evidence behind AI release recommendations. Engineering owns remediation, the privacy and youth-safety advisor guides policy and escalation requirements, and organizational leadership approves material risk decisions.

AI safety lead responsibilities

Build a risk-based evaluation plan for correctness, grounding, instructional usefulness, uncertainty, fairness, privacy leakage, age appropriateness, and safe character boundaries.
Create representative scenarios and calibrated review rubrics using synthetic or approved, minimized data. Include different reading levels, learner needs, and multi-turn conversations.
Test direct and indirect prompt injection, misleading retrieved content, cross-user information exposure, unauthorized tool actions, jailbreaks, and failures after interruptions or provider errors.
Design youth-safety tests for harmful advice, bullying, manipulation, emotional dependency, and appropriate responses to urgent concerns. Coordinate escalation expectations with the youth-safety owner.
Build a lightweight automated harness that compares prompt, model, retrieval, and tool changes against a versioned regression set.
Combine automated results with human review. Track severity, affected scenarios, evaluation limits, and differences across relevant learner groups.
Write reproducible issue reports, work with owners on fixes, and verify that remediation addresses the failure rather than simply hiding the test case.
Define release criteria, monitoring signals, incident-learning practices, and a sustainable evaluation playbook. Conduct adversarial testing only within approved systems and written boundaries.

Early deliverables

First 30 days: map the highest-priority risks, establish evaluation rubrics, and create an initial representative test set.
By 60 days: run an automated and human-reviewed evaluation cycle, document significant failures, and agree on release criteria with leadership.
By 90 days: integrate repeatable checks into the release workflow and deliver a readiness report with remediation owners, retest evidence, and remaining limitations.

AI safety lead required qualifications

Demonstrated work in AI evaluation, adversarial testing, software quality, application security, trust and safety, or a closely related discipline.
Practical understanding of LLM applications, retrieval, tool calling, prompt layers, and common system-level failures.
Ability to build or maintain evaluations with Python, JavaScript, or TypeScript and use APIs, structured test data, and version control.
Sound judgment about metrics, sampling, rubric calibration, and the limits of model-based grading.
Clear technical writing, discretion with sensitive material, and the ability to challenge a release decision constructively.

AI safety lead helpful experience

How we work

Partner with engineering, product, learning design, UX, privacy, and security contributors. Use minimized evidence in reports and maintain clear stop conditions for sensitive or high-risk testing.

Why volunteer with iLevelUP

Contribute to a nonprofit education product at a stage when your decisions can shape the learner experience.
Own a meaningful body of work and collaborate across education, design, engineering, and student support.
Develop work samples and share approved contributions in your portfolio while respecting project and learner confidentiality.

Time commitment and location

Fully remote and unpaid. Contribute at least 10 hours per week for 12 months, with collaboration scheduled at mutually workable times.

How to apply

Apply for the AI Evaluation, Red Team & Safety Lead volunteer role. Use the Believe in Me volunteer application. Share your résumé or LinkedIn profile, relevant work samples, weekly availability, and why this role interests you. Share only authorized materials and remove personal, confidential, or proprietary information.

Include a sanitized evaluation report, test harness, threat model, or case study. Describe an AI failure you made reproducible, how you judged its severity, and how you verified the fix.

We welcome applicants from varied backgrounds and pathways. Demonstrated capability, judgment, and dependable collaboration matter more than a particular degree. If you need an accommodation for the application or interview process, contact us through iLevelUP’s contact page.

Role details

Role: AI Evaluation, Red Team & Safety Lead
Program: iLevelUP, a Believe in Me initiative
Employment type: Unpaid volunteer
Workplace: Fully remote
Applicant eligibility: United States
Team: App Development Team
Company address: 107 Spring Street, 2nd Floor, Seattle, WA 98104
Job ID: 1010

Fully remote and unpaid. Contribute at least 10 hours per week for 12 months, with collaboration scheduled at mutually workable times.