Red Team Content Safety VLM AI Trainer, $60-$70/hour

Project Overview

Join a growing community of professionals advancing the next wave of AI. As an AI Trainer, you’ll play a hands-on role by analyzing and providing feedback on data to improve LLM performance, helping ensure that the next generation of AI technology is accurate and trustworthy.

We are seeking a skilled AI Red Team and Content Safety Professional to work as a project consultant in our AI Labor Marketplace. This is not a full-time employment position — you will be engaged as an expert project consultant on a contract basis.

Location: U.S.-based experts only

Engagement: Part-time, project-based expert evaluation work

Work Type: Remote

Project Summary

Contributors will conduct adversarial safety evaluations of vision-language models using realistic image-and-text challenges. The work focuses on identifying content-safety failures, including harmful model assistance and unnecessary refusals of legitimate requests.

Task authors will design and execute tests in which the image is essential, capture and annotate actual model responses, document evidence supporting their assessments, and revise submissions based on independent review. Reviewers will independently evaluate the challenge design, model behavior, and supporting evidence.

Testing occurs in a controlled environment and may involve sensitive or disturbing content.

Consultant Engagement Terms

This is a project-based consultant role. Consultants will be paid on a per-project basis; hourly rates are estimates based on anticipated completion time. Consultants control their own schedule, provide their own tools, and may simultaneously provide services to other vendors or employers (subject to those vendors’ allowances).

Responsibilities

Contributors will:

• Design realistic adversarial image-and-text challenges that test VLM content-safety behavior.
• Create tests in which visual information is essential to the evaluation.
• Execute challenges and accurately capture and annotate actual model responses.
• Evaluate potential harmful assistance as well as unnecessary refusals of legitimate requests.
• Document evidence supporting each assessment clearly enough for independent review.
• Apply contextual safety judgment to complex or ambiguous model behavior.
• Revise author submissions based on reviewer feedback.
• For reviewer assignments, independently assess test quality, model behavior, evidence, and whether conclusions are sufficiently supported.

Expected Outcomes

• High-quality, reproducible VLM safety evaluation tasks.
• Image-and-text challenges that meaningfully test the intended safety behavior.
• Accurate records and annotations of observed model responses.
• Clear, evidence-based assessments suitable for independent review.
• Responsive revisions where reviewer feedback identifies issues.
• Independent reviewer decisions with clear feedback when revisions are required.

Qualifications

• Hands-on AI red-teaming experience is strongly preferred, particularly experience involving images, multimodal systems, or vision-language models.
• Relevant backgrounds may also include cybersecurity red teaming, trust and safety, content policy, policy research, fraud or abuse investigation, communications, rhetoric, linguistics, conversation design, or behavioral research.
• Demonstrated adversarial thinking, contextual judgment, careful experimentation, and clear written communication.
• Ability to distinguish meaningful safety failures from ambiguous or acceptable model behavior.
• Relevant specialist expertise for assignments involving areas such as privacy, hate and bias, manipulation, wellbeing, accessibility, regulated advice, or civic integrity.
• Reviewers must have sufficient relevant depth to assess submitted work independently.
• No fixed degree, years-of-experience, target-company, or employer-pedigree requirement.
• Coding is not a general requirement; specialist assignments may require relevant technical competence.
• Demonstrated ability will be assessed through practical screening.

LinkedIn Data Security and Privacy Notice

https://legal.linkedin.com/content/dam/legal/Global-Data-Privacy-Notice_LI_TalentCommunity.pdf