Simple math formula predicts when AI chatbots will go rogue

Researchers propose a simple mathematical formula that predicts when small on-device AI chatbots will start producing harmful answers. The risk is especially acute offline, where there is almost no safety oversight.

• The formula predicts when a chatbot will start giving harmful answers
• Small on-device chatbots have little safety oversight
• Offline mode is especially risky with no filters or moderation
• Risks include self-harm, financial loss and extremist content