Remote AI Safety Red Teaming Specialist

OpenTrain AI; Österreich; Remote; EUR 19.000 - 26.000; Teilzeit

Benefits dieser Stelle

Remote work

Flexible hours

Contractor status

Zusammenfassung

OpenTrain AI is recruiting an AI Safety Red Teaming Expert for text-based contractor work focused on adversarial testing of conversational AI models and agents. You will investigate jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation, then produce structured human data that identifies vulnerabilities and systemic risks.

The work may involve sensitive subjects, including bias, misinformation, and harmful behavior.

Qualifikationen

• Native fluency in English and Telugu; strong language quality is essential.
• Ability to evaluate accuracy, completeness, and appropriateness of outputs.
• Understanding of jailbreaks, prompt injections, misuse cases, and multi-turn manipulation.
• Ability to classify vulnerabilities with standardized taxonomies and playbooks.
• Strong written communication and clear reasoning.
• Ability to notice subtle errors, inconsistencies, gaps, and risks.
• Adaptability across projects and evaluation contexts.

Aufgaben

• Test model behavior across varied scenarios and assess safety of outputs.
• Probe jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation.
• Review outputs for accuracy, completeness, and potential harm.
• Annotate failures and classify vulnerabilities using taxonomies and benchmarks.
• Flag systemic risks and identify gaps in evaluation coverage.
• Document reproducible attack cases, reports, and datasets.
• Expand testing across scenarios that automated tests may overlook.

Kenntnisse

English fluency

Telugu fluency

Judgment

Language quality

Attention to detail

Adaptability

Jobbeschreibung

OpenTrain AI is recruiting an AI Safety Red Teaming Expert for text-based contractor work focused on adversarial testing of conversational AI models and agents. You will investigate jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation, then produce structured human data that identifies vulnerabilities and systemic risks.

The work may involve sensitive subjects, including bias, misinformation, and harmful behavior.