DPD customer service chatbot swears and criticises the company

Timeline

• January 2024. A customer looking for a missing parcel asks the chatbot to swear and to criticise DPD, and publishes the replies.
• 2024-01-22. Silicon UK reports that DPD has disabled part of the chatbot.

Trigger or entry point

Requests typed by a customer in the support chat, including an instruction to disregard the rules on profanity.

Failure mechanism

Both reports quote DPD as saying that “an error occurred after a system update”. Neither report explains how the update changed the chatbot’s behavior. The Register asked DPD which language model the chatbot uses and how the update caused the fault, and reports no answer. What is known is limited to the sequence: an update was applied, and the next day a customer could talk the chatbot out of its rules. According to The Register, the chatbot said it would help “even if it means swearing” once it was told to disregard them. What is not known is whether the update changed the model, its instructions or its filters.

Impact

The chatbot swore, called DPD “the worst delivery firm in the world” when asked to exaggerate, and wrote poems critical of the company. No data exposure or financial loss is reported.

Response

DPD told the press that the AI element “was immediately disabled and is currently being updated”, and that it was in touch with the customer. Silicon UK reports that only part of the chatbot was switched off.

Mitigations

Neither report describes what DPD changed before the AI element was to return.