AI Safety Experts — English & Thai

$24–35/hrrate as listed

Mercor is staffing a bilingual red team that probes conversational AI models in English and Thai for jailbreaks, prompt injections, bias exploitation, and multi-turn manipulation. Work is remote, text-based, and paid at an observed $24–35/hr, with sensitive-topic projects clearly flagged in advance and optional.

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 spend your sessions trying to break a model rather than use it. That means constructing adversarial prompts in both English and Thai, chaining multi-turn conversations that erode a model's refusals, testing prompt-injection paths through tool use or retrieved content, and probing where a safety policy holds in one language but collapses in the other. Cross-lingual gaps are a large part of why this role is bilingual: a guardrail trained mostly on English data often behaves differently when the same request arrives in Thai, in transliterated Thai, or in a code-switched mix.

When an attack lands, the artifact matters more than the exploit. You annotate the failure, classify it against the project's vulnerability taxonomy, judge severity, and write it up so an engineer who has never seen your conversation can reproduce it. Consistency is enforced through playbooks and benchmarks — the goal is a dataset customers can act on, not a collection of one-off clever tricks.

What the screen looks for

Verifiable red teaming history — adversarial ML work, penetration testing, trust-and-safety abuse analysis, or socio-technical probing, described with specifics rather than tool names.
Genuine native-level fluency in both languages, including register, slang, and the ways Thai speakers actually phrase evasive or coded requests.
Structured thinking under follow-up. Expect the interviewer to push on how you'd systematically cover an attack surface, not just what your best jailbreak was.
Write-up quality. Reproducibility, clear severity reasoning, and plain-language risk explanation carry real weight.

Logistics

Fully remote and asynchronous, contractor terms, with contributors typically choosing their own hours against weekly volume commitments. Work is entirely text-based. Higher-sensitivity workstreams — bias, misinformation, harmful behaviors — are opt-in, with topics disclosed before exposure and wellness resources provided. Project availability shifts with customer demand, so hours can vary week to week; the $24–35/hr band reflects rates observed on this listing and is not a guarantee.

What they look for

Native fluency in both English and Thai
Prior red teaming experience (adversarial ML, cybersecurity, or trust & safety probing)
Familiarity with LLM attack classes: jailbreaks, prompt injection, multi-turn manipulation
Structured testing against taxonomies, benchmarks, or playbooks
Clear reproducible technical writing for mixed technical/non-technical readers
Comfort working with sensitive content under documented guidelines

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.

Is English your native language, or equivalent native fluency acquired through childhood immersion?
Native fluency in English · required
Is Thai your native language, or equivalent native fluency acquired through childhood immersion?
Native fluency in Thai · required
Do you have at least 6 months of hands-on experience in AI red teaming, adversarial ML testing, cybersecurity testing, or socio-technical probing?
Red teaming or adversarial testing experience · required
Are you able to work remotely and have reliable internet access?
Remote work access · required
Are you willing to engage substantively with sensitive topics such as bias, misinformation, and harmful behaviors as part of your red teaming work?
Comfortable with sensitive content · preferred

0 of 5 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 the candidate has genuine adversarial testing experience (not just theoretical knowledge) and can explain their process reproducibly rather than through luck or surface-level poking.
A strong answer: Specific example from cybersecurity, ML research, security testing, or similar work where the candidate executed a structured attack or probing method, identified a real flaw others missed, and documented it in a way that drove action or remediation.
Where people lose it: Vague or secondhand stories ('I read about jailbreaks'), or discoveries that were accidental rather than the result of deliberate methodology—reveals someone who tests randomly, not systematically.