AI Safety Experts — English & Punjabi

$20–22/hrrate as listed

Red team conversational AI models in English and Punjabi, 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 Punjabi, running multi-turn conversations designed to erode refusal behavior, testing prompt injection against agentic tool use, and probing where a model's safety training thins out when the language shifts. Punjabi coverage matters here: safeguards tuned on English data often fail differently in Gurmukhi or Shahmukhi script, in transliterated Punjabi, or in code-mixed Punjabi-English speech, and part of your job is finding exactly where.

When an attack lands, the deliverable is not the exploit — it's the write-up. You classify the failure against a provided taxonomy, document the reproduction steps, note whether the behavior is stable across reruns, and explain severity in terms a customer's safety team can act on. Consistency with the playbook matters more than cleverness; a brilliant one-off jailbreak that nobody can reproduce is worth less than a well-characterized failure mode.

Some projects touch bias, misinformation, harassment, or other harmful-content categories. Topics are disclosed before you see any content, participation in higher-sensitivity work is optional, and wellness resources are available.

What the screen looks for

Genuine bilingual depth. Native-level Punjabi that holds up under register shifts, dialect variation, and script switching — not conversational familiarity.
Prior adversarial work. AI red teaming, penetration testing, trust-and-safety abuse analysis, or comparable socio-technical probing, with specifics you can describe.
Structured method. Whether you work from a taxonomy or benchmark rather than improvising, and whether you can articulate your coverage strategy.
Written clarity. Your reports are the product. Expect the screen to weigh how precisely you describe a failure and its severity.

Logistics

Fully remote and asynchronous. Work is text-based and delivered as project batches, so hours are largely self-scheduled, though individual customer engagements may carry turnaround windows. Volume fluctuates with project demand — treat this as variable contract work rather than a steady weekly load. Observed pay is $20–22/hr; rates are set per project and not guaranteed.

What they look for

Native fluency in both English and Punjabi, including script and dialect variation
Prior red teaming experience — AI adversarial testing, cybersecurity, or trust-and-safety probing
Familiarity with LLM attack surfaces: jailbreaks, prompt injection, multi-turn manipulation
Clear technical writing and reproducible documentation of findings
Ability to apply taxonomies and benchmarks consistently across large batches
Comfort working with sensitive content categories including bias, misinformation, and harmful behavior

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 both English and Punjabi?
Native fluency in English and Punjabi · required
Do you have documented experience in red teaming, adversarial machine learning, cybersecurity testing, or socio-technical probing of AI systems or security contexts?
Prior red teaming or adversarial AI experience · required
Are you able to work fully remote?
Remote work location · preferred

0 of 3 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: This tests whether you have genuine red teaming or adversarial ML experience, not just awareness of the field. It reveals your actual attack surface knowledge and whether you can articulate a reproducible finding.
A strong answer: Describe a concrete case (jailbreak, prompt injection, bias exploitation, or security bypass) with the system context, your specific approach or taxonomy you applied, what failed, and the artifact or report you produced—demonstrating you can move beyond random testing to structured, documentable work.
Where people lose it: Candidates often describe general security awareness or ethical concerns without naming a specific system, technique, or output they created. Vague answers ('I once tested a chatbot') signal no hands-on red teaming experience.