Public story · 2026-07-02 · high
Meta Sent Fake Teen Crisis Prompts to Rival Chatbots
Codenamed Cannes, the operation used hundreds of contractors to target ChatGPT, Gemini and Character.AI, with one August round topping 45,000 prompts.
Why now: Wired's account published July 1, a day before it landed in this week's briefing.
Story
Meta hired hundreds of contractors to pose as teenagers and flood ChatGPT, Gemini and Character.AI with crisis prompts, according to Wired.
The prompts referenced suicide, self-harm and drug use, the same categories crisis-response systems are built to catch. Wired put a number on just one round: more than 45,000 prompts in August 2025, aimed at how three rival products handle real users in danger.
Wired reported the effort ran under the internal codename Cannes. Contractors built accounts flagged as belonging to minors, then used them to test how each competitor's chatbot responded to self-harm and substance-related conversations.
Meta hasn't confirmed the operation, and Wired's report doesn't say why it ran Cannes or what happened to the data it collected.
My bet is none of the three targeted companies admit what they found in their own logs. Doing so means admitting they can't tell Meta's fake crisis accounts from the real ones. Watch their next safety disclosure for a mention, or the silence where one should be.
Wired's account published July 1, a day before it landed in this week's briefing.
Related stories
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source domain / Earlier coverage
June 2026 was the biggest model launch month ever, and ChatGPT just dropped below 50% share
Both cover ChatGPT, Gemini; reported by the same outlet (unrot.co); earlier ChatGPT coverage from 2026-06-21.
Shared entities / Shared topic / Earlier coverage
Claude web traffic up 306% in a quarter as ChatGPT's share falls to 54.7%.
Both cover ChatGPT, Gemini; overlapping topics (chatbot, chatgpt); earlier ChatGPT coverage from 2026-06-08.
UK Brings AI Chatbots Under Online Safety Act with 10% Revenue Fines
Both cover ChatGPT, Gemini; overlapping topics (chatbot, chatgpt); earlier ChatGPT coverage from 2026-02-16.
Shared entities / Earlier coverage / Tension
GPT-5.4 Mini ($0.75/M) and Nano ($0.20/M) — Purpose-Built for Subagent Workloads
Both cover ChatGPT, Gemini; earlier ChatGPT coverage from 2026-03-17; pushes against this story (vs).
MCP Ecosystem Crosses 10,000 Servers Under Linux Foundation Governance
Both cover ChatGPT, Gemini; earlier ChatGPT coverage from 2026-02-23; pushes against this story (vs).
Shared entity: ChatGPT / Shared topic / Earlier coverage / Tension
Prosecutors entered ChatGPT logs as evidence in the Palisades fire arson trial.
Both cover ChatGPT; overlapping topics (chatbot, chatgpt); earlier ChatGPT coverage from 2026-06-28.
Shared entity: Meta / Shared topic / Earlier coverage
Meta's AI support chatbot became a seven-week password-reset backdoor. 20,225+ Instagram accounts hijacked.
Both cover Meta; overlapping topics (account, chatbot, meta); earlier Meta coverage from 2026-06-08.
Shared entities / Earlier coverage
Noam Shazeer left Google for OpenAI, two years after Google paid $2.7B to get him
Both cover Character, Gemini; earlier Character coverage from 2026-06-21.
Source trail
Entities
Claim evidence
- Meta Sent Fake Teen Crisis Prompts to Rival Chatbots
Provenance
- Canonical issue
- 2026-07-02
- AI generated
- yes
- Story unit
- 2026-07-02-meta-ran-a-covert-operation-sending-crisis-prompts-to-rival-chatbots-via-fake-teen-accounts
- Labels
- source-backed, canonical briefing excerpt