What Happened
A frustrated customer successfully jailbroke DPD's chatbot, getting it to write a poem calling DPD "the worst delivery firm in the world" with profanity. The chatbot also agreed it was useless and criticized the company.
Impact
Viral embarrassment, DPD had to disable chatbot temporarily. Demonstrated how easily customer-facing AI can be manipulated to damage brand reputation.
Cost: Millions in negative PR, temporary loss of chatbot service
How to Prevent This
- Implement robust jailbreak detection and prevention
- Add content filtering for all outputs (not just inputs)
- Test chatbots with adversarial prompting before launch
- Create clear boundaries on what chatbot can discuss