My support chatbot scored 0.92. It was also lying to customers.

Twelve of my thirteen test prompts scored a perfect 1.00. Two independent evaluation runs agreed:...