Heroines, not heroin: Facebook page returns after AI flagged it for drugs (www.theguardian.com)

A Gloucestershire photography charity, Hundred Heroines: Women in Photography Today, had its Facebook group taken down twice in 2025 after Meta’s automated moderation tools flagged the name as promoting the class‑A opioid “heroin” (the charity’s name uses “heroines”). After more than a month of appeals following a September takedown — and no clear human response or apology — the page was restored. The small organisation, founded in 2020 with a physical collection of about 8,000 items, says roughly 75% of its visitors come via Facebook, so the repeated removals had a “devastating” impact on outreach and audience access. The case highlights wider issues in AI content moderation: scale-driven reliance on automated detection can produce false positives from simple keyword or pattern‑matching, opaque decisioning, and insufficient human‑in‑the‑loop review. Meta has said it uses AI centrally to detect drug-related content amid heightened enforcement after the US opioid crisis, but errors — and prior mass suspensions tied to a technical error — show the risks for nonprofits, cultural groups and other edge cases. The incident underscores the need for more contextual NLP, clearer appeal channels, explainability and targeted human oversight so moderation systems don’t inadvertently silence legitimate communities.