AI Safety Experts — English & Odia

$20–22/hrrate as listed

Red team conversational AI models in Odia and English, probing for jailbreaks, prompt injections, bias exploitation, and multi-turn manipulation. Mercor lists this project at an observed $20–22/hr, fully remote and text-based.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

What the work involves

You will spend most of your time in adversarial conversation with AI models — writing attack prompts, escalating across multiple turns, and looking for the point where safety behavior degrades. Work is split between English and Odia, and a substantial part of the value here is testing whether guardrails that hold in English hold equally in Odia: transliteration tricks, code-switching, culturally specific framings, script-level manipulation, and low-resource-language blind spots.

When you find a failure, the artifact matters as much as the attack. You will annotate what broke, classify the vulnerability against the project taxonomy, note reproducibility conditions, and write it up so a customer engineering team can act on it. Consistency is enforced through playbooks and benchmarks — this is structured probing, not freeform hacking.

What the platform screens for

Genuine native fluency in both languages, including register, idiom, and regional variation in Odia — expect to be tested, not just asked.
Prior adversarial experience of some kind: AI red teaming, penetration testing, trust and safety, disinformation research, or abuse analysis.
Systematic thinking. Screeners probe whether you work from a threat model and coverage plan or just improvise.
Write-up quality. Vague findings fail; reproducible, well-scoped reports pass.

Logistics and content

Remote, asynchronous, and text-only. Contributors typically set their own hours against weekly throughput expectations; steady part-time availability is generally more useful to the project than sporadic bursts. The work involves reviewing model outputs touching bias, misinformation, and harmful behavior. Mercor states that higher-sensitivity streams are opt-in, topics are disclosed before exposure, and wellness resources are provided. Pay is stated as observed on the platform at $20–22/hr and is not guaranteed for any individual contract.

What they look for

Native fluency in both English and Odia (spoken and written)
Prior red teaming experience — AI adversarial testing, cybersecurity, or socio-technical probing
Working knowledge of LLM attack surfaces: jailbreaks, prompt injection, multi-turn manipulation
Structured documentation: reproducible bug reports, vulnerability classification, taxonomy adherence
Comfort reviewing sensitive content including bias, misinformation, and harmful outputs
Familiarity with adversarial ML or trust-and-safety frameworks (nice to have)

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Are you a native English speaker with daily fluency in both written and spoken English?
Native English fluency · required
Are you a native Odia speaker with daily fluency in both written and spoken Odia?
Native Odia fluency · required
Do you have prior experience in red teaming, adversarial AI testing, cybersecurity penetration testing, or socio-technical probing?
Prior red teaming or adversarial testing experience · preferred
Are you able to work fully remote, asynchronously, and meet Mercor's project deadlines?
Remote work eligibility · required

0 of 4 answered

Included with this listing

Interview prep — 10 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether candidate has genuine red teaming or adversarial testing experience, and whether they can articulate structured methodology rather than luck-based discovery.
A strong answer: Describe a specific vulnerability (in AI, security, or a complex system), explain the hypothesis or framework that guided your search, and show how you reproduced and documented the finding in a way others could act on it.
Where people lose it: Talking about finding a bug in production without explaining the methodology, or describing curiosity without any framework—'I just tried things until something broke' signals you'll generate noise, not reproducible vulnerabilities.