NYU study finds AI chatbots give unsafe abuse survivor advice

AI chatbots frequently provide unsafe or incomplete guidance to survivors of technology-enabled abuse, according to new research from NYU Tandon School of Engineering. The study evaluated multiple leading models on their responses to high-stakes scenarios involving scams, hacked accounts, and harassment. Researchers found that automated advice often lacked necessary safety protocols or omitted critical context required for victim protection. This investigation is part of a broader effort to assess AI reliability in sensitive situations where errors could cause direct harm. The findings highlight significant risks in deploying general-purpose language models as substitutes for professional crisis support. While specific model names were not disclosed in the submission, the research underscores systemic deficiencies in current safety alignment techniques. Industry stakeholders face renewed pressure to implement specialized safeguards for vulnerable user demographics before expanding AI into mental health or legal advisory roles.