This is a fully remote position, open to applicants in United States.

📋 Description

• Conduct red-team operations on conversational AI models and agents by executing jailbreaks, prompt injections, misuse scenarios, bias exploitation, and multi-turn manipulations.

• Create human data through the annotation of failures, classification of vulnerabilities, and identification of systemic risks.

• Utilize taxonomies, benchmarks, and playbooks to ensure consistent testing protocols.

• Develop reproducible reports, datasets, and attack scenarios for clients.

• Identify vulnerabilities that automated testing may overlook.

• Broaden evaluation coverage and mitigate unexpected issues during production.

• Assist Mercor customers in enhancing the safety, robustness, and trustworthiness of their AI systems.

⛳️ Requirements

• Fluent or native proficiency in both English and Danish is essential.

• Previous experience in red teaming, particularly in AI adversarial contexts, cybersecurity, or socio-technical probing.

• Capability to challenge AI systems adversarially and push them to their limits.

• Familiarity with frameworks, taxonomies, benchmarks, or playbooks for organized testing methodologies.

• Skill in articulating risks clearly to both technical and non-technical audiences.

• Flexibility to adapt across various projects and clients.

• Preferred skills include: adversarial ML, jailbreak datasets, prompt injection, RLHF/DPO attacks, model extraction, penetration testing, exploit development, reverse engineering, harassment/disinformation probing, abuse analysis, conversational AI testing, psychology, acting, or creative adversarial writing.

• Must be eligible to work without H1-B or STEM OPT sponsorship.