A new position paper argues that current AI/ML research on deepfakes is misaligned with the prevalent issue of AI-generated non-consensual intimate imagery (AIG-NCII). The paper highlights that existing research primarily focuses on epistemic harms, such as authenticity and fraud, rather than the dignity harms experienced by subjects of AIG-NCII. It suggests that technical interventions often overlook the subject-centric nature of these harms, and that knowing an image is synthetic does not alleviate the damage to individuals. The authors recommend realigning AI safety research to include subject-centric threat models and establish partnerships with experts in sexual violence prevention. AI
IMPACT Current AI safety research on deepfakes needs to shift focus to address the harms of AI-generated non-consensual intimate imagery.
RANK_REASON Academic paper discussing AI safety and ethical concerns. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
