Remote AI training and data labeling jobs

AI safety testing in Telugu and English: provoke and document jailbreaks, bias and harmful output from chat models, and judge the quality of their Telugu answers. Evaluation skill matters more than security experience. Remote hourly contract at $16–22/hour, weekly pay; H-1B and STEM OPT excluded.

Odia and English AI safety work: probe chat models for jailbreaks, bias and harmful output and judge whether their Odia answers hold up. One of the lowest-resource languages in the family and the busiest Indian variant, with 88 hired this month. Remote hourly contract, $16–22/hour.

Red-team AI models in European and other non-Brazilian Portuguese plus English: jailbreaks, prompt injection, bias and manipulation, logged as reproducible safety data. Brazilian Portuguese is explicitly excluded. Remote hourly contract at $29–45/hour, paid weekly.

Review AI-narrated audiobooks in European Portuguese and flag where the synthetic narrator goes wrong: dropped or added words, bad pronunciation, misread numbers and abbreviations, unnatural rhythm. For native speakers from Portugal who listen to audiobooks. Remote hourly contract, around 20 hours a week, $15–20/hour.

Native Japanese speakers write prompts on sensitive subjects, classify conversations and flag adversarial phrasing so AI models behave safely in Japanese. Business English and a bachelor's (in progress is fine). Part-time remote at $48–52/hour; Japan or East Asia preferred, not required; 8 hires this month.

Red-team conversational AI models in Norwegian and English: attempt jailbreaks, prompt injections and multi-turn manipulation, then document what broke. Prior red-teaming, security or adversarial testing experience expected. Remote hourly contract at $48–62/hour, paid weekly.

Native Croatian speakers write prompts on sensitive subjects, classify prompts and conversations, and flag adversarial phrasing to harden AI models in Croatian. Bachelor's (in progress is fine) and business English. Part-time remote at $38–42/hour; Croatia or Eastern Europe preferred, not required.

AI safety work in Kannada and English: test chat models for jailbreaks, bias and harmful output, and judge whether their Kannada answers are accurate and appropriate. Unlike the European variants, prior red-teaming is not listed as a requirement. Remote hourly contract, $16–22/hour, weekly pay.

Red-team AI models in Dutch and English: jailbreaks, prompt injection, bias exploitation and multi-turn manipulation, written up as reproducible attack cases and labelled data. Native Dutch plus prior adversarial or security experience. Remote hourly contract at $48–62/hour, paid weekly.

Take recorded one-on-one Spanish lessons with experienced tutors so a language-learning platform can capture authentic learner conversations. For native English speakers with a degree and at least a year of Spanish study. About 7 hours a week for roughly three weeks, $30–37/hour, remote.

Red-team AI chat models and agents in Thai and English, then document every jailbreak, injection or biased answer as reproducible safety data. Native Thai required, plus prior adversarial or security experience. Remote hourly contract at $24–35/hour, paid weekly; 205 hired this month.

Native Thai speakers write prompts on sensitive subjects, classify conversations and flag adversarial phrasing to make AI models safer in Thai. Business English and a bachelor's (in progress is fine) required. Part-time remote at $18–22/hour; Southeast Asia preferred, not required.

The highest-paid generalist role in Mercor's bilingual AI safety series: native Norwegian speakers write sensitive-topic prompts, classify conversations and flag adversarial phrasing. Bachelor's (in progress is fine) and business English. Part-time remote at $58–62/hour; Norway preferred, not required.

Audit Brazilian Portuguese speech data for Amazon's Sonic project: check transcripts against audio with fixed error codes and pass/fail verdicts, and fix word-level timestamp alignment. Native Portuguese as spoken in Brazil is a hard requirement; rationales are written in English. Remote hourly contract at $25/hour.

Quality-check AI-narrated audiobooks in German (Germany): pinpoint mispronounced, missing or extra words, wrongly read numbers and abbreviations, unnatural prosody and audio artefacts, and judge the overall listen. For native German audiobook listeners. Remote hourly contract, about 20 hours a week, $15–20/hour.

Native or near-native English editors, proofreaders and linguists evaluate LLM-generated English, rewrite it to a high stylistic standard, annotate grammatical and stylistic features, and give feedback that guides model training at a leading AI lab. Remote, flexible hours, $50/hour.

Evaluate AI-narrated audiobooks in Dutch (Netherlands): mark each mispronounced, skipped or added word, misread number, awkward intonation and audio glitch, then rate the overall listen. For native Dutch speakers who are regular audiobook listeners. Remote hourly contract, about 20 hours a week, $15–20/hour.

Listen to AI-narrated audiobooks in Brazilian Portuguese and tag where the synthetic narrator slips: mispronunciations, skipped or extra words, wrong readings of numbers and abbreviations, flat or odd intonation. For native Brazilian audiobook listeners. Remote hourly contract, around 20 hours a week, $15–20/hour.

Test AI models for safety failures in Finnish and English: jailbreaks, prompt injection, bias and multi-turn manipulation, logged as structured red-team data. For native Finnish speakers with prior adversarial, security or abuse-analysis experience. Remote hourly contract, $48–62/hour.

Swedish and English red-teaming of AI models: jailbreaks, injected instructions, bias and multi-turn manipulation, written up as reproducible attack cases. The busiest listing in this family, with 243 hires this month. Remote hourly contract at $48–62/hour, paid weekly.

Test AI models in Gujarati and English for jailbreaks, bias and harmful answers, and judge whether their Gujarati is accurate and appropriate. Evaluation judgment is the core requirement. Open to the Gujarati diaspora, with no residence rule published. Remote hourly contract at $16–22/hour; 56 hired this month.

Listen to AI-narrated Italian audiobooks and log every slip in the synthetic voice: skipped or mispronounced words, misread numbers, odd intonation, glitches. For native Italian speakers who actually listen to audiobooks. Remote hourly contract, about 20 hours a week, $15–20/hour.

Probe AI chat models and agents for safety failures in Danish and English, then turn each failure into labelled, reproducible red-team data. For native Danish speakers with a security, adversarial ML or abuse-analysis background. Remote hourly contract, $48–62/hour, weekly pay.

Write sensitive-topic prompts in Ukrainian and classify prompts and conversations for an AI safety project, flagging adversarial phrasing as you go. Native Ukrainian, business English and a bachelor's (finished or in progress). Part-time remote at $38–42/hour; Eastern Europe preferred, not required.