Trending PR Campaigns
DistressBench: Testing Chatbots in Mental-Health Crises
Scale AI worked with 19 clinicians and crisis counsellors to create 718 realistic crisis conversations, then tested 25 leading AI models on how they responded. The hook was a new benchmark, DistressBench, revealing that chatbots often recognised distress but failed to direct users to useful help—giving TIME a timely, evidence-led story about a high-stakes gap in AI safety.
Why it worked: It paired exclusive, clinician-authored benchmark data with a clear failure rate (35%) across 25 marquee models, turning a vague safety worry into hard, newsy accountability.
The lesson: Consider commissioning domain experts to stress-test a popular tool, publish a simple metric of failure, and offer exclusive first access.
Coverage over time
How this story spread across the media
| Publication | Headline | Date |
|---|---|---|
| nature.com | Preparing AI chatbots to respond to patient distress and suicidality in high-risk healthcare settings | npj Digital Medicine | Tuesday 22nd September, 2026 |
| time.com | AI Chatbots Often Fail To Help Users in Mental-Health Crisis | Friday 9th October, 2026 |
| stfrancisdesalespaducah.org | AI Chatbots in Crisis: $5 Million in Funding to Test and Improve Safety (2026) | Friday 16th October, 2026 |