AI Safety Expert – English and Bengali (Remote Contract)
Mercor is seeking an English- and Bengali-speaking AI safety red teamer to identify vulnerabilities in conversational AI systems. The role involves producing reproducible attack cases and human-generated data for AI labs training frontier models.
Description
- Test conversational AI models and agents using jailbreaks, prompt injection, misuse scenarios, bias exploitation, and multi-turn manipulation
- Create human-generated data by annotating failures, categorizing vulnerabilities, and identifying systemic risks
- Use established taxonomies, benchmarks, and playbooks to keep evaluations consistent
- Deliver reproducible reports, datasets, and attack cases for customers
- Assess AI responses involving bias, misinformation, harmful behavior, and other sensitive subjects
- Broaden evaluation coverage and contribute to stronger customer AI systems
Requirements
- Native or fluent proficiency in both English and Bengali
- Sound judgment when evaluating language and content, including the ability to determine whether AI responses are accurate, complete, and appropriate and explain the assessment
- Strong attention to subtle errors, inconsistencies, and omissions
- Reliable adherence to testing guidelines and quality standards
- Clear communication of reasoning to both technical and non-technical audiences
- Willingness to adapt across projects, task formats, and customers
- Ability to work as an independent contractor
- Must not require H-1B or STEM OPT sponsorship
- Preferred: adversarial machine learning experience with jailbreak datasets, prompt injection, RLHF/DPO attacks, or model extraction
- Preferred: cybersecurity experience in penetration testing, exploit development, or reverse engineering
- Preferred: socio-technical risk experience in harassment or disinformation probing, abuse analysis, or conversational AI testing
- Preferred: psychology, acting, or writing experience that supports unconventional adversarial thinking
Benefits
- Work fully remotely
- Set a flexible schedule and work according to your own availability
- Receive weekly payments through Stripe or Wise
- Choose whether to participate in higher-sensitivity projects
- Access clear guidance and wellness resources for sensitive-content assignments
- Receive competitive compensation
- Build experience in human-data-driven AI red teaming
- Collaborate with leading AI researchers
- Earn referral payments of up to $90 for each successful referral