Findings from transparency notices on AI companion apps: October 2025 (non-periodic)

Key Findings

• — None of the four providers had robust age assurance; all relied on app store ratings and/or self-declaration at signup, leaving children able to reach adult spaces and features
• — Chai, Chub AI and Nomi did not direct users to support or help services when self-harm was detected in user prompts
• — Chub AI and Nomi were not checking model inputs and outputs (and Chai not checking outputs) across all text, image and video models for CSEA, self-harm material or pornography
• — Nomi and Chub AI had no staff dedicated to trust and safety or moderation; Chub AI and Nomi did not red-team across all models used in their services
• — Neither Chai nor Nomi stated they reported detected CSEA material to an enforcement authority or NCMEC
• — Post-notice changes: Chub AI geo-blocked Australia; Character.AI introduced age assurance and sentiment analysis; Chai restricted free companion chat and added real-time prompt redirection; Nomi committed to improved CSEA/self-harm detection
• — Survey: 79% of Australian children 10-17 had used an AI companion or assistant; 54% of users used them for companion-type purposes including mental health advice (20%) and chatting about feelings (22%)

Methodology Notes

Compulsory transparency notices under Australia's Basic Online Safety Expectations (Online Safety Act 2021), requiring four providers to detail safety systems for the period 1 Jul-30 Sep 2025; supplemented by a demographically representative 2026 survey of 1,950 Australian children aged 10-17. Findings reflect provider self-reports in response to notice questions across eight themes.

Sources