Skip to main content
Peer-reviewed Authoritative — Peer-reviewed venues, standards bodies, regulators, official government publications

“CHAT WITH ME below as if I were a human”: Insights from Auditing AI Chatbots for Survivors of Domestic Violence

The first global characterisation of the chatbots that domestic-violence services put in front of survivors. The authors identify 36 such chatbots across 22 countries, document what privacy terms they publish, and run scenario-based audits on a purposive subset of nine LLM-powered web chatbots in English or Spanish. Most have no chatbot-specific privacy policy and few ask the follow-up questions that would contextualise a survivor's need.

Read at doi.org (opens the original source in a new tab)

Publisher

ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Brown University; University of Maryland

Published

25 Jun 2026

Added

today

Key Findings

  • 36 domestic-violence chatbots identified across 22 countries.
  • 16 of the 36 had no privacy policy at all, and only 7 had one specifically addressing the chatbot.
  • Scenario-based audits were run on a purposive subset of nine LLM-powered web chatbots operating in English or Spanish.
  • Only a few of the audited chatbots asked follow-up questions to contextualise the user's situation before responding.
  • The audit surfaces a gap between the safety-critical population these services address and the absence of basic data-handling commitments.

Methodology Notes

Ecosystem identification plus a purposive audit subset of nine; scenario-based rather than deployed-traffic evidence; English and Spanish only. FAccT 2026 pp. 1398-1423, DOI 10.1145/3805689.3812260, published 2026-06-25; dl.acm.org is unreachable from this box, so metadata (title, authors, affiliations, venue, pages, date) was verified through the Crossref record.

Topics

social relational harmVulnerable usershuman ai interaction safetyGuardrails & moderationPrivacy & data protectionStandards & governance

Authors

Julio Poveda, Alyssa Lanter, Isabela Antoniuk, Michelle L. Mazurek, Diana Freed

Tags

facct-2026domestic-violenceauditprivacy-policybrownmaryland

Cite This

APA

Julio Poveda et al. (2026). “CHAT WITH ME below as if I were a human”: Insights from Auditing AI Chatbots for Survivors of Domestic Violence. ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Brown University; University of Maryland. https://doi.org/10.1145/3805689.3812260