Remote AI Safety Specialist — Adversarial ML & Risk Testing

mercor; San Francisco (CA); Híbrido; GBP 17.000 - 23.000; A tiempo parcial

Descripción de la vacante

Mercor is seeking AI Safety Experts to join our remote team to stress-test conversational AI models, identify jailbreaks, prompt injections, and bias exploitation across Odia and English. This contract role offers flexible, asynchronous work and exposure to cutting-edge adversarial testing.

You will annotate failures, classify vulnerabilities, document reproducible reports, and collaborate with researchers to improve model safety.

Formación

• Native fluency in English and Odia.
• Strong judgment about language and content accuracy.
• Rigorous attention to subtle errors and inconsistencies.
• Ability to work to guidelines and quality standards.
• Clear communication with technical and non-technical audiences.
• Adaptability across projects, task types, and customers.

Responsabilidades

• Red team conversational AI models and agents to identify jailbreaks, prompt injections, misuse cases, and bias exploitation.
• Generate high-quality human data by annotating failures, classifying vulnerabilities, and flagging systemic risks.
• Apply structure by following taxonomies, benchmarks, and playbooks to maintain consistent testing.
• Document reproducibly by producing reports, datasets, and attack cases that customers can act on.
• Work independently and asynchronously to meet deadlines while improving AI model performance.

Conocimientos

English

Odia

Attention to detail

Communication

Independent work

Adaptability

Technical writing

Descripción del empleo

Mercor is seeking AI Safety Experts to join our remote team to stress-test conversational AI models, identify jailbreaks, prompt injections, and bias exploitation across Odia and English. This contract role offers flexible, asynchronous work and exposure to cutting-edge adversarial testing.

You will annotate failures, classify vulnerabilities, document reproducible reports, and collaborate with researchers to improve model safety.