Overview
- Bereaved parents say their children confided suicidal thoughts to the chatbot in lengthy private exchanges discovered after their deaths.
- OpenAI says 0.15% of roughly 800 million weekly users trigger conversations with suicide indicators, totaling more than a million instances each week.
- The company reports updated models that better detect distress and cut undesired responses by at least 65%, with internal tests meeting support goals 91% of the time.
- Experts and advocates, including Suicide Prevention Australia, argue self-regulation is insufficient and warn chatbot design can validate harmful thinking and foster dependence.
- Policy pressure is rising as a Hawley–Blumenthal bill would restrict youth access with age checks, recurring nonhuman disclosures, and criminal penalties, while U.S. and Australian regulators step up oversight.