Job Description:
We are seeking highly analytical and detail-oriented professionals with hands-on experience in Red Teaming, Prompt Evaluation, and AI/LLM Quality Assurance. The adecuado candidate will help us rigorously test and evaluate AI-generated content to identify vulnerabilities, assess risks, and ensure compliance with safety, ethical, and quality standards.
Key Responsibilities:
- Conduct Red Teaming exercises to identify adversarial, harmful, or unsafe outputs from large language models (LLMs).
- Evaluate and stress-test AI prompts across multiple domains (e.g., finance, healthcare, security) to uncover potential failure modes.
- Develop and apply test cases to assess accuracy, bias, toxicity, hallucinations, and misuse potential in AI-generated responses.
- Collaborate with data scientists, safety researchers, and prompt engineers to report risks and suggest mitigations.
- Perform manual QA and content validation across model versions, ensuring factual consistency, coherence, and guideline adherence.
- Create evaluation frameworks and scoring rubrics for prompt performance and safety compliance.
- Document findings, edge cases, and vulnerability reports with high clarity and structure.
Requirements:
- Proven experience in AI red teaming, LLM safety testing, or adversarial prompt design.
- Familiarity with prompt engineering, NLP tasks, and ethical considerations in generative AI.
- Strong background in Quality Assurance, content review, or test case development for AI/ML systems.
- Understanding of LLM behaviors, failure modes, and model evaluation metrics.
- Excellent critical thinking, pattern recognition, and analytical writing skills.
- Ability to work independently, follow detailed evaluation protocols, and meet tight deadlines.
Preferred Qualifications:
- Prior work with teams like OpenAI, Anthropic, Google DeepMind, or other LLM safety initiatives.
- Experience in risk assessment, red team security testing, or AI policy & governance.
Background in linguistics, psychology, or computational ethics is a plus.