Ethical Concerns and Organizational Image As Factors to AI Adoption In Social Work Among Students, Faculty, Staff, and Practicum Supervisors

Introduction: As to adopting AI large language models (AI LLMs) to social work practice, qualitative studies suggest that ethical concerns and potential violations of workplace rules and norms are primary barriers, yet quantitative evidence supporting this observation remains limited. Guided by the Technology Acceptance Model 2 (TAM2; Venkatesh & Davis, 2004), this study examines how demographic factors, personal values and ethical concerns related to AI (e.g., AI guilt), and perceived image enhancement within one’s organization (e.g., image) explain variability in the adoption of in social work.

Methods: Respondents (N = 121) were recruited in 2026 from a social work program. The sample included 64 students (43.5% BSW; 56.5% MSW; 67.7% online; 32.3% in-person), 36 faculty and staff with an average of 9.97 years in higher education(SD = 6.61 years), and 21 practicum supervisors of 12.6 years (SD=6.0 years) working as a social worker. Demographic variables included age, race/ethnicity (non-Hispanic White vs. others), and English confidence (extremely confident vs. less than extremely confident). Consistent with TAM2, measures included perceived ease of use, subjective norms, voluntariness, image, and organizational support. Ethical concerns were assessed using AI guilt that measures emotional discomfort experienced when using AI LLM for creative activities such as writing. AI LLM adoption was measured using intention to use (four items) as well as the AI Social Work Task Application Scale (AISWTAS) which assesses willingness (1–5) to use AI across 24 tasks related to documentation, communication, and access to information. Multiple regression analyses were conducted for adjusted analyses with a significance level of .05.

Results: In bivariate analyses, AISWTAS scores were positively associated with perceived ease to use (r=.33, p<.001), subjective norms(r=.36, p<.001), image (r=.42, p<.001), and voluntariness (r =.36, p< .001), and negatively associated with AI guilt (r=−.49, p< .001). Intention to use showed a similar pattern, with positive associations for perceived ease to use (r = .33, p<.001), subjective norms (r=.38, p<.001), image (r=.46, p<.001), voluntariness (r = .31, p<.001), and organizational support (r=.23, p=.011), and a negative association with AI guilt (r=−.60, p<.001). In adjusted analyses, AISWTAS scores (adjusted R² = .422) were significantly predicted by AI guilt (β=−.33, p<.001), image (β = .28, p<.001), and voluntariness (β=.23, p=.007). Intention to use (adjusted R²=.563) was similarly explained by AI guilt (β=−.45, p<.001), image (β=.274, p<.001), and voluntariness (β=.17, p=.027).

Discussion and Conclusions: Findings indicate personal values and ethical concerns as well as image within their organization associated with AI LLM’s application to social work jobs. AI guilt emerged as a particularly important construct, reflecting emotional discomfort with delegating professional tasks to AI LLM. These findings highlight the need for open dialogue and structured guidance on ethical, responsible, and value-coherent AI use in social work before implementing more advanced technical support within agencies and educational programs.