AI Safety Red Teamer — English & Urdu Fluency

About the role

We're building a global red team of human data experts to help make AI systems safer and more trustworthy. In this role, you'll act as an adversarial tester — probing AI model outputs with carefully crafted inputs to surface vulnerabilities, biases, and potential harms before they reach real users. Your work will directly shape the safety guardrails of AI systems used by millions.

This is a remote, text-based project ideal for detail-oriented bilingual professionals who are comfortable engaging critically with sensitive and nuanced content, including topics related to bias, misinformation, and harmful behaviors.

What you'll do

• Design and submit adversarial prompts intended to expose weaknesses, biases, or unsafe behaviors in AI model outputs
• Review and evaluate AI-generated responses for safety, accuracy, and appropriateness
• Document findings clearly, categorizing types of vulnerabilities or failure modes discovered
• Collaborate with a distributed team to refine red-teaming methodologies and share insights
• Provide written feedback and structured annotations in both English and Urdu
• Maintain consistency and rigor when evaluating sensitive or controversial content

Requirements

• Native or near-native fluency in both English and Urdu (written and comprehension)
• Strong critical thinking skills and comfort engaging with sensitive subject matter (bias, misinformation, harmful content)
• Excellent written communication and attention to detail
• Ability to work independently in a fully remote, text-based environment
• Reliable internet access and availability for consistent, ongoing contribution
• Prior experience with content moderation, linguistics, AI evaluation, or trust & safety work is a plus

Compensation

This role pays $20–$22 per hour, based on experience and performance. Engagement is remote and flexible, with opportunities for ongoing work as the project scales.