Skip to content
Cendo Labs
How it worksWhat we testLive demoPricingBuildersGuidesFree taster

Incident write-up · 8 October 2026

A delivery firm's chatbot swore at a customer and mocked the company.

By , founder of Cendo Labs

Short answer: In January 2024, a customer trying to track a missing parcel got so little help from DPD's chatbot that he started testing it instead. He got it to swear, write a poem about how bad DPD was, and call it the worst delivery firm in the world. DPD said a system update had caused the behaviour and switched off part of the chatbot. A chatbot that can be coaxed off-brand will be, and the screenshots travel.

What happened

A musician was trying to find a missing parcel through DPD's online chat in the UK. The chatbot couldn't help him, so he started asking it to do other things. It swore when asked to, wrote a poem about how useless DPD was, and described DPD as the worst delivery firm in the world.

He shared the conversation online and it was widely reported. DPD said an error after a system update had caused the behaviour, disabled the AI part of the chatbot, and began updating it.

Why it matters for your business

  • A frustrated customer is a tester. When a chatbot can't solve the problem, people poke at it, and some share what they find.
  • Updates change behaviour. According to DPD, the problem appeared after a system update. A chatbot that behaved well last month may not this month.
  • Off-brand is still public. Nobody lost money here, but the company's name sat next to the chatbot's words in news articles.

How to prevent it

  1. Give the bot clear limits on tone: no swearing, no criticising the business or competitors, no role-play, whatever the customer asks.
  2. Make sure it can hand over. Much of this started because the customer couldn't get his problem solved or reach a person.
  3. Re-test after every update, including updates you didn't make yourself, such as a new version of your chatbot platform or its AI model.

How Cendo Labs tests for this

Our AI tester tries "ignore your previous instructions" tricks, role-play games and deliberately rude, provoking messages, and checks whether your bot hands a frustrated customer to a real person. The bot fails if it drops its role, loses its temper or produces content that would embarrass the business. See "ignore your instructions", angry customer and asking for a person.

Sources

Want to know how your own chatbot does?

Test my chatbot free