AI Safety Experts — English & Bengali

$20–22/hrrate as listed

Red team conversational AI models in English and Bengali, probing for jailbreaks, prompt injections, bias exploitation, and multi-turn manipulation. Work is text-based, remote, and asynchronous, with observed pay in the $20–22/hr range.

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 on purpose. That means constructing adversarial prompts in both English and Bengali, running multi-turn manipulation attempts, testing prompt injection through tool or document context, and probing where safety training thins out under translation, code-switching, or transliterated Bengali script. When an attack lands, the deliverable is not the screenshot — it is a reproducible artifact: the exact prompt chain, the failure classification against the project taxonomy, a severity judgment, and a note on why the model failed rather than just that it did.

Much of the day-to-day is annotation and classification discipline. You will work inside taxonomies and playbooks rather than freestyling, tag failure modes consistently across sessions, and flag patterns that look systemic rather than one-off. Some tasks touch sensitive content — bias, misinformation, harassment vectors, harmful instruction requests. Mercor states that higher-sensitivity queues are opt-in, topics are disclosed before exposure, and wellness resources are available.

What the screen looks for

Genuine bilingual depth. Native-level English and Bengali, including register shifts, regional variation, and Romanized Bengali — not translation-app fluency.
Prior adversarial work. AI red teaming, penetration testing, exploit development, trust-and-safety abuse analysis, or socio-technical probing. Expect follow-ups asking you to walk through a specific attack you ran and why it worked.
Structured method over clever one-offs. Interviewers probe whether you can generalize an attack into a test class and describe coverage.
Clear written reasoning. Your report is the product; ambiguity in your write-up is a defect.

Logistics

Fully remote and asynchronous, contractor engagement, work claimed from queues rather than assigned on a fixed schedule. Volume fluctuates by customer project, so treat it as variable-hour rather than steady full-time. Observed pay for this listing is $20–22/hr; rates are set per project and are not guaranteed. Expect a calibration period with sample tasks and reviewer feedback before full-rate work opens up.

What they look for

Native fluency in both English and Bengali, including Romanized/transliterated Bengali
Prior red teaming experience — AI adversarial testing, cybersecurity, or trust-and-safety abuse analysis
Working knowledge of jailbreaks, prompt injection, and multi-turn manipulation techniques
Ability to document attacks reproducibly with clear severity and failure classification
Comfort applying a fixed taxonomy or benchmark consistently across long annotation sessions
Willingness to review sensitive content (bias, misinformation, harmful behavior) in text form

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 a language you grew up speaking natively (either as a first language or simultaneously from early childhood), and do you use it regularly in your daily life or professional work?
Native English fluency · required
Is Bengali a language you grew up speaking natively (either as a first language or simultaneously from early childhood), and do you use it regularly in your daily life or professional work?
Native Bengali fluency · required
Do you have prior hands-on experience in red teaming, adversarial AI testing, cybersecurity penetration testing, or similar probing work—even if informal or self-directed?
Red teaming or adversarial testing experience · preferred
Are you available to work remotely, and can you reliably work on this project at the stated rate of $20–22 per hour?
Availability for remote work · required

0 of 4 answered

Included with this listing

Interview prep — 12 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 red team or adversarial testing experience, and whether they use structured methods (frameworks, taxonomies, playbooks) or just intuition and luck.
A strong answer: The candidate names a specific vulnerability (in AI, security, or a comparable domain), explains the structured methodology or framework they used to uncover it, and describes how they documented or reproduced it so others could act on it.
Where people lose it: Describing a single lucky find or fixing a bug they stumbled upon, without evidence of systematic probing, repeatable methodology, or understanding of why the vulnerability matters—leaving the screener uncertain whether they can do this reliably.