Probe conversational AI for jailbreaks, prompt injections, bias, misinformation, and other safety failures in English and Bengali. This remote freelance role pays $20-$22 per hour.
Generative AI & RLHF
100% Remote Hourly · $20–$22/hr
$20–$22/hr
Compensation
Worldwide
Eligibility
Expert
Experience
Jul 13, 2026
Posted
Open worldwide
Interested in this role?
Create a free OpenTrain account and apply in minutes.
OpenTrain is the #1 platform for finding and building careers in AI training and data labeling. OpenTrain AI hires and contracts contributors for specialized projects, helping you build a durable portfolio of work teaching and evaluating AI systems.
Create a free OpenTrain account and apply in minutes.
Build a profile that showcases your AI training and safety evaluation experience.
Discover opportunities that match your skills and support long-term growth in the field.
About AI Safety Training
AI training is the human side of building modern artificial intelligence. Safety specialists test model behavior, document weaknesses, and provide structured feedback that helps AI systems become more reliable, secure, and responsible.
Remote work completed with a computer and internet connection.
Use creative thinking and disciplined evaluation to shape cutting-edge conversational AI.
Contribute human-generated data, attack cases, and risk analysis that automated tests may miss.
The Role
OpenTrain AI is seeking an expert Conversational AI Safety Red Team Specialist to probe conversational models and agents with adversarial inputs. The work is text-based and combines creative adversarial thinking with structured evaluation, documentation, and risk reporting.
You will investigate weaknesses involving jailbreaks, prompt injections, misuse, bias, misinformation, harmful behavior, and multi-turn manipulation. The role requires native fluency in English and Bengali and experience with AI adversarial testing, cybersecurity, or socio-technical probing.
Role type: Remote freelance contractor
Experience level: Expert
Pay: $20-$22 per hour
Schedule: Part time, with a default commitment of 40 hours per week; the structured requirement is 20+ hours per week
Languages: Native English and Bengali
Higher-sensitivity topics are optional and come with topic guidance and wellness resources before exposure.
What You'll Do
You will evaluate conversational AI systems systematically, turning discovered failures into clear, reproducible findings. Your work will support datasets, benchmarks, and reports that make safety risks actionable for AI system teams.
Test conversational models and agents against jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation.
Review outputs involving sensitive topics and identify failures, vulnerabilities, and systemic risks.
Annotate failures and classify vulnerabilities using taxonomies, benchmarks, and established playbooks.
Produce reproducible reports, datasets, and attack cases.
Expand evaluation coverage by uncovering weaknesses that automated testing may miss.
Required Qualifications
This is an expert-level role requiring prior experience applying structured methods to adversarial AI or related safety testing. You should be able to communicate both technical and societal risks clearly and reproducibly.
Prior red teaming experience involving AI adversarial testing, cybersecurity, or socio-technical probing.
Native fluency in both English and Bengali.
Ability to identify jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation.
Experience applying taxonomies, benchmarks, or playbooks to structured model evaluation.
Clear written and verbal communication of technical and non-technical AI risks.
Curiosity, creativity, adaptability, and sound judgment when exploring model behavior.
Helpful Experience
The following backgrounds may help you approach conversational AI weaknesses from unconventional angles. They are useful supporting experience in addition to the required red teaming and structured evaluation skills.
Adversarial machine learning or jailbreak datasets
Prompt injection, RLHF or DPO attacks, or model extraction
Penetration testing, exploit development, or reverse engineering
Harassment, misinformation, or abuse analysis
Conversational AI testing
Psychology, acting, or writing experience that supports creative adversarial thinking
How to Apply
Apply through OpenTrain AI to be considered for this remote freelance opportunity. The work is designed for contributors who can commit substantial weekly availability while maintaining careful, consistent documentation of model behavior.
Creating an OpenTrain account is free. Your profile can help demonstrate credible AI training experience, discover future opportunities, and develop a long-term portfolio in a fast-growing field.
Set up or update your OpenTrain profile.
Highlight your red teaming, AI safety, cybersecurity, or socio-technical testing experience.
Show your English and Bengali fluency and structured evaluation skills.
Apply and complete any role-specific evaluation steps requested by OpenTrain AI.
Test conversational AI for jailbreaks, prompt injections, bias, misuse, and multi-turn manipulation. This expert contract role offers 20+ hours weekly at $48-$62 per hour for native English and Dutch speakers.
Probe conversational AI for jailbreaks, prompt injections, bias exploitation, and manipulation as an expert red team contractor. Work worldwide for $48 to $62 per hour, 20+ hours weekly, using English and Danish.
Help improve conversational AI by uncovering jailbreaks, prompt injections, bias risks, and other vulnerabilities. This remote contractor role offers 20+ hours per week at $29-$45 per hour for native English and Portuguese speakers.