Moderators call for AI controls after Reddit Answers suggests heroin for pain relief (www.engadget.com)

A Reddit moderator flagged Reddit Answers after the AI suggested dangerous medical advice—telling a user to stop prescriptions and take high‑dose kratom, giving mixed neonatal fever guidance, and even proposing heroin for chronic pain. Mods across health-focused subreddits reported they couldn't hide, disable, or flag those AI responses within their communities, raising alarms because the tool appeared publicly in conversations where incorrect or harmful guidance could reach vulnerable users. Reddit said it updated Reddit Answers to stop showing “Related Answers” on sensitive topics and that the feature already excludes private, quarantined and NSFW communities and “some mature topics,” but it hasn’t clarified which topics are blocked or given moderators controls. The episode highlights core AI-moderation problems: model hallucinations and unsafe completions around medical content, weak topic detection/guardrails, and a lack of per‑community opt‑out or moderation tooling. For the AI/ML community this underlines the need for stricter domain‑specific safety layers, human‑in‑the‑loop verification for medical queries, finer-grained content filters and transparency about classifier rules to prevent dangerous recommendations at scale.