"Better Ask for Forgiveness than Permission": Practices and Policies of AI Disclosure in Freelance Work

Paper Title

"Better Ask for Forgiveness than Permission": Practices and Policies of AI Disclosure in Freelance Work

Publication Info

• Topic area: AI use and disclosure practices in freelance work.
• Keywords: AI disclosure, freelance platforms, client-worker trust, AI policies, transparency, generative AI, AI governance, trust dynamics, AI adoption, policy design.

Background and Problem

• Problem / challenge: Freelancers and clients have misaligned expectations regarding AI use and disclosure. Workers often adopt passive disclosure practices, while clients expect proactive transparency. Existing AI policies are vague, leading to misunderstandings and inconsistent practices.
• Significance: Misaligned expectations can erode trust, disrupt client-worker relationships, and hinder the development of effective AI governance in freelance and other decentralized work contexts.
• Motivation and related work: Prior research highlights the growing use of AI in professional settings and its impact on trust and perceptions of quality. However, little is known about how freelancers navigate AI use and disclosure in the absence of clear organizational policies, making this study critical for understanding these dynamics.

Solution

• Proposed approach: A three-stage study to investigate AI use and disclosure practices among freelancers and clients, including interviews, surveys, and policy evaluations.
• Novelty:
  1. Identification of five AI disclosure practices among freelancers.
  2. Analysis of misalignments between workers’ and clients’ expectations regarding AI use and disclosure.
  3. Examination of how client AI policies shape workers’ interpretations and behaviors.
  4. Recommendations for proportional, stage-aware, and scenario-driven AI policies.
• Procedure and key techniques:
  1. Conducted interviews with 41 freelancers to explore AI use and disclosure practices.
  2. Surveyed 100 freelancers and 145 clients to compare perspectives on AI use and disclosure.
  3. Presented client AI policies to 100 additional freelancers to evaluate their interpretations and intended behaviors.

Results

• Concrete findings:
  • 78.83% of freelancers use AI, often considering it essential for productivity.
  • Passive disclosure (disclosing AI use only when asked) is the most common practice among freelancers (39%).
  • Clients expect more proactive disclosure but are less confident in detecting AI use (M = 3.11 on a 5-point scale).
  • Workers misinterpret AI policies, particularly under partial-permission policies, leading to over- or underestimation of allowed AI use.
• Advantage over baselines: The study provides a nuanced understanding of AI use and disclosure in freelance work, highlighting gaps in existing policies and offering actionable recommendations for improvement.
• Experiments / evaluation:
  • Study 1.1: Interviews with 41 freelancers.
  • Study 1.2: Survey of 100 freelancers on AI use and disclosure.
  • Study 2: Survey of 145 clients on AI expectations and policies.
  • Study 3: Survey of 100 freelancers evaluating client AI policies.
• Limitations and future work:
  • Focused on freelance platforms, limiting generalizability to other work settings.
  • Future research should explore AI disclosure in other precarious labor markets and test proportional disclosure frameworks experimentally.

Summary

This study investigates the practices and policies of AI disclosure in freelance work, revealing significant misalignments between freelancers’ passive disclosure practices and clients’ expectations for proactive transparency. Through interviews and surveys, the study identifies five disclosure strategies and highlights the challenges posed by vague or binary AI policies. The findings underscore the need for proportional, stage-aware, and scenario-driven policies to bridge expectation gaps and foster trust. Recommendations include clearer policy templates, disclosure ladders, and participatory policy design to support both workers and clients. These insights have implications for AI governance in decentralized and institutionalized work contexts alike.