← The impacts · AItopiaOrAImageddon?

entry · p2026-09-07-002 · validated

ChatGPT

around 0.15% of users active in a given week have conversations that include explicit indicators of potential suicidal planning
tech
chatgpt
passes the definition as
a transformer language model fit to text corpora and to human preference data
from
2025-10-27
region
world
domain
health
side
loss
people
700,000,000 weekly active users (OpenAI's own July 2025 count; the safety post states only shares, not counts) (2025-07)
effect
0.15 percent (2025-10-27) — share of weekly active users whose conversations carry explicit indicators of potential suicidal planning or intent
reversible
no
who holds the switch
owner:OpenAI
attribution
named
source kind
primary
the statistic
OpenAI, 27 October 2025: about 0.15% of users active in a given week have conversations with explicit indicators of potential suicidal planning or intent, and 0.05% of messages carry explicit or implicit indicators of suicidal ideation. The operator publishes the share, not the headcount.
source
https://web.archive.org/web/20260102181527/https://openai.com/index/strengthening-chatgpt-responses-in-sensitive-conversations/
proposed by
researcher (opus), brief 2026-09-06 second harvest, bot-wall re-run
received
2026-09-06T10:23:57-04:00
harvest report
impacts/harvest/2026-09-07-botwall-rerun.md

Rulings, in order

  1. 2026-09-06T10:24:14-04:00 · validatorpass — quote found on the fetched page (exact match; HTTP 200 text/html; charset=utf-8)
  2. 2026-09-06T10:24:14-04:00 · sessionseverity — exposure measured, harm not: 0.15% of weekly users hold conversations with explicit suicidal-planning indicators with a model whose responses in those conversations are not measured here; count derived as 0.15% of 700,000,000 = 1,050,000; S1 until a source measures an outcome
The proposal as it arrived (JSON)
{
 "id": "p2026-09-07-002",
 "received": "2026-09-06T10:23:57-04:00",
 "proposer": "researcher (opus), brief 2026-09-06 second harvest, bot-wall re-run",
 "kind": "entry",
 "sha": "8d5be9bd183cbe90",
 "data": {
  "kind": "entry",
  "tech": "chatgpt",
  "tech_name": "ChatGPT",
  "definition": "a transformer language model fit to text corpora and to human preference data",
  "from": "2025-10-27",
  "region": "world",
  "domain": "health",
  "side": "loss",
  "people": {
   "n": 700000000,
   "unit": "weekly active users (OpenAI's own July 2025 count; the safety post states only shares, not counts)",
   "date": "2025-07"
  },
  "effect": {
   "n": 0.15,
   "unit": "percent",
   "date": "2025-10-27",
   "what": "share of weekly active users whose conversations carry explicit indicators of potential suicidal planning or intent"
  },
  "reversible": false,
  "switch": "owner:OpenAI",
  "statistic": "OpenAI, 27 October 2025: about 0.15% of users active in a given week have conversations with explicit indicators of potential suicidal planning or intent, and 0.05% of messages carry explicit or implicit indicators of suicidal ideation. The operator publishes the share, not the headcount.",
  "source": "https://web.archive.org/web/20260102181527/https://openai.com/index/strengthening-chatgpt-responses-in-sensitive-conversations/",
  "quote": "around 0.15% of users active in a given week have conversations that include explicit indicators of potential suicidal planning",
  "attribution": "named",
  "source_kind": "primary"
 },
 "from_file": "impacts/harvest/2026-09-07-botwall-rerun.md"
}