AI Safety Experts — English & Vietnamese

$17–25/hrrate as listed

Red team conversational AI models in English and Vietnamese, probing for jailbreaks, prompt injections, bias exploitation, and multi-turn manipulation. You document each successful attack as a reproducible artifact — taxonomy-classified, annotated, and usable by the customer engineering team.

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

The work

This is adversarial testing of conversational AI, done bilingually. You'll run structured attack sessions against models and agents — single-turn jailbreaks, prompt injection through tool or document context, misuse framings, bias elicitation, and slow multi-turn manipulation where the model concedes ground gradually. A large part of the value here is cross-lingual: safety training frequently generalizes poorly from English to Vietnamese, and attacks that fail in one language often succeed when translated, code-switched, or reframed through Vietnamese cultural and political context. Native fluency in both languages is a hard requirement, not a preference.

Every finding becomes an artifact. You annotate the failure, classify it against the project taxonomy, note reproduction steps and the model's behavior across retries, and write it up so a non-specialist can understand why it matters. Random one-off hacks are less useful here than coverage against a benchmark — the platform is explicit that structured probing beats improvisation.

What the screen looks for

Verifiable prior red teaming. AI adversarial work, penetration testing, exploit development, trust-and-safety abuse analysis, or socio-technical probing. Expect follow-ups on specific attacks you've run and what the system did.
Genuine bilingual depth. Screens for Vietnamese language roles typically test register, idiom, and regional variation — not just conversational ability. Translation-tool fluency does not pass.
Judgment under ambiguity. Whether a borderline output counts as a failure, how you'd rate severity, and when a refusal is over-cautious rather than safe.
Sensitive-content readiness. The work touches bias, misinformation, self-harm, and harassment material. Topics are disclosed before exposure and higher-sensitivity projects are opt-in, but you should be honest with yourself about fit.

Logistics

Fully remote and largely asynchronous, with work assigned in project batches that rotate across customers. Observed pay for this listing runs $17–25/hr, with placement in the band typically reflecting prior security or adversarial-ML background. Hours are flexible and self-scheduled; contributors commonly report 10–25 hours per week, though volume depends on active customer projects rather than a fixed commitment. Expect calibration materials, a taxonomy document, and periodic guideline updates you're responsible for keeping current with.

What they look for

Native fluency in both English and Vietnamese
Prior red teaming experience — AI adversarial work, cybersecurity, or socio-technical probing
Familiarity with jailbreak techniques, prompt injection, and multi-turn manipulation patterns
Structured testing against taxonomies, benchmarks, or attack playbooks
Clear written documentation and reproducible reporting
Willingness to work with sensitive content (bias, misinformation, harmful behaviors)

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 speaker of English, and are you a native speaker of Vietnamese, or did you grow up using both as primary languages?
Native fluency in English and Vietnamese · required
Have you previously worked in red teaming, adversarial testing of AI systems, cybersecurity penetration testing, or socio-technical risk probing (such as testing conversational AI for bias, misinformation, or abuse)?
Prior red teaming, adversarial ML, cybersecurity, or socio-technical risk experience · required
Have you used a formal framework, taxonomy, benchmark, or structured playbook to guide security testing or adversarial work—for example, MITRE, OWASP, a custom red teaming checklist, or a model evaluation benchmark?
Ability to work with structured taxonomies or frameworks · preferred
Are you available to work as a remote contractor on a flexible schedule?
Remote work availability · preferred

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 the candidate has genuine adversarial experience (AI jailbreaking, prompt injection, cybersecurity, or socio-technical probing) and can articulate the gap between automated testing and human insight.
A strong answer: The answer must name a concrete system (an AI model, application, network, or social process), describe a specific vulnerability the candidate uncovered through probing rather than static analysis, and explain the reasoning or technique that revealed it—not luck.
Where people lose it: Candidates claim red teaming experience but describe only standard security audits, compliance checks, or theoretical knowledge rather than hands-on adversarial work; or they name a vulnerability but cannot explain how they discovered it or why it was hard to find.