Meta secretly tested ChatGPT, Gemini, and Character.AI with thousands of minor-perspective crisis prompts

· Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI Ethics & Safety · Depth: Fundamental Awareness, quick

Summary

Meta secretly conducted extensive safety testing on rival AI chatbots, including ChatGPT, Gemini, and Character.AI, using thousands of crisis prompts from the perspective of minors. Internally named "Cannes," the project involved hundreds of contractors from Covalen creating fake underage accounts to send prompts about self-harm, eating disorders, and drugs. Active through April 2026, one August 2025 round saw over 45,000 prompts. Meta defended this as industry-standard safety testing, stating the collected responses were not used for training its own AI models. However, the tested companies, unaware of the activity, indicated potential terms of service violations, with Character.AI confirming a breach.

Key takeaway

For AI developers and product managers designing or deploying conversational AI, you must prioritize transparent and ethical safety testing protocols, especially when potential interactions involve minors. Ensure your terms of service explicitly address such testing scenarios and consider robust age verification mechanisms to mitigate risks. This incident underscores the critical need for industry-wide standards to prevent unauthorized data collection and potential legal or reputational damage.

Key insights

Covertly testing AI chatbots with minor-perspective crisis prompts raises significant ethical, legal, and terms of service concerns.

Principles

Topics

Best for: CTO, Executive, VP of Engineering/Data, Tech Journalist, AI Ethicist, Legal Professional

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.