AI Safety Expert – English, Gujarati

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

📋 Description

• Conduct red-team assessments of conversational AI models and agents by employing jailbreaks, prompt injections, misuse scenarios, bias exploitation, and multi-turn manipulation techniques.

• Create human data by annotating failures, categorizing vulnerabilities, and identifying systemic risks.

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

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

• Evaluate AI outputs related to sensitive issues such as bias, misinformation, or harmful behaviors.

• Identify vulnerabilities that automated tests may overlook.

• Broaden evaluation coverage to minimize unexpected outcomes in production.

• Enhance customer AI systems through adversarial testing.

⛳️ Requirements

• Native fluency in both English and Gujarati is essential.

• Strong judgment regarding language and content is necessary.

• Capable of evaluating whether AI responses are accurate, complete, and appropriate, with the ability to articulate the reasoning behind assessments.

• Skilled at noticing subtle errors, inconsistencies, and gaps in AI outputs.

• Ability to consistently adhere to guidelines and maintain quality standards.

• Proficient in conveying reasoning clearly to both technical and non-technical audiences.

• Adaptable to various projects, task types, and client needs.

• Must operate as an independent contractor.

• Should not require H1-B or STEM OPT sponsorship.

• Nice-to-have: experience in adversarial machine learning, including knowledge of jailbreak datasets, prompt injection, RLHF/DPO attacks, or model extraction.

• Nice-to-have: background in cybersecurity, such as penetration testing, exploit development, or reverse engineering.

• Nice-to-have: experience in socio-technical risk, including harassment/disinformation probing, abuse analysis, or testing conversational AI.

• Nice-to-have: expertise in psychology, acting, or writing that supports unconventional adversarial thinking.

🏝️ Benefits

• Fully remote position.

• Flexible working hours; tasks can be completed according to your own schedule.

• Weekly payments through Stripe or Wise based on services provided.

• Project durations may vary based on needs and performance, allowing for extensions, reductions, or early conclusions.

• Participation in higher-sensitivity projects is optional.

• Clear guidelines and wellness resources are available for work involving sensitive content.

• Reasonable accommodations are provided upon request.

• Opportunity to gain experience in human data-driven AI red teaming.

• Chance to contribute to making AI systems more robust, safe, and trustworthy.

• Competitive compensation.

• Collaboration with leading researchers in the field.