AI Safety Experts: English & Tamil

What you'll do

Tamil has a split that trips AI models up: the formal written language differs markedly from spoken Tamil, and online users mix both with English and with romanised "Tanglish". It is also spread across several countries (India, Sri Lanka, Singapore, Malaysia and a large diaspora), each with its own sensitivities. A model tuned mainly on English is unlikely to handle all of that safely, and Mercor's AI safety project wants native Tamil speakers to find out where it fails.

The ad's task list is red-team work:

• Push chat models and agents with jailbreaks, prompt injections, misuse cases, bias exploitation and multi-turn manipulation
• Annotate and classify failures, and flag systemic risks
• Follow taxonomies, benchmarks and playbooks
• Write reproducible reports, datasets and attack cases

The candidate profile for Tamil, however, emphasises evaluation judgment over security experience: can you tell if an AI response is accurate, complete and appropriate, catch subtle errors, and explain your reasoning? That is a lower barrier than the "prior red teaming experience" asked of the European variants.

All tasks are text. Sensitive topics (bias, misinformation, harmful behaviour) are part of the work; you are told the topics first, and higher-sensitivity projects are optional with guidelines and wellness resources.

Who fits

• Native fluency in Tamil and English
• Good judgment on the accuracy, completeness and appropriateness of AI answers
• Close attention to detail and consistency with guidelines
• Clear written explanations
• Comfort moving between projects and task types

Extras the ad welcomes: adversarial ML, cybersecurity, harassment and disinformation analysis, psychology, acting or writing. No degree or minimum years are stated.

What it pays

$16–22 per hour, Mercor's published band for all the Indian-language variants and Urdu. For the same template, Thai pays $24–35 and the Nordic and Dutch versions $48–62.

The ad states weekly payment via Stripe or Wise. H-1B and STEM OPT candidates are not supported.

Worth knowing

Good:

• 57 hires on this listing this month, so it is actively staffing
• Evaluation-focused profile that does not demand security experience
• No residence requirement published, which matters for a language spoken across several countries
• Weekly US-dollar payouts
• Sensitive projects are opt-in, topics announced first

Less good:

• Lowest pay band in this family
• The tasks are still adversarial, and some content is distressing
• No weekly hours stated; projects can end early
• Contractor terms, clients unnamed

About this listing

Posted by Mercor as a remote hourly contract, read from the Mercor site on 24 September 2026. Pay, requirements, payment terms and the visa exclusion are as published. Mercor states the work involves no confidential information from any employer, client or institution. See Mercor.