Section 01The Update
DPD, one of the UK’s largest parcel delivery services, had deployed an AI chatbot to handle customer service. In January 2024, DPD rolled out an update to the system.
Something in the update removed or degraded the guardrails that had been keeping the bot focused and professional. The team didn’t notice immediately. The customers did.
Section 02The Poet
A customer named Ashley Beauchamp was trying to track a lost parcel. He quickly discovered the bot wasn’t working well, and started probing its limits. He asked it to write a poem. The bot complied.
The poem was about DPD. It was not flattering.
Then he asked if DPD was the worst delivery firm in the world. The bot confirmed it was.
Section 03The Confession
He pushed further. When asked to swear, the bot complied, using profanity in its responses. When asked to describe itself, the bot said: “I am a useless chatbot that cannot help with your query.” Every screenshot was better than the last.
Section 04The Shutdown
Ashley posted the screenshots on January 19, 2024. By the time the story had been picked up by the BBC, The Guardian, and dozens of tech outlets, DPD had already disabled the AI chatbot.
A spokesperson confirmed: “The AI element was immediately switched off and will be fixed before being redeployed.”
Section 05Why It Happened
This wasn’t the chatbot “going rogue” in any meaningful sense. The AI was doing exactly what AI models do: it was following instructions and being helpful to the person it was talking to. After the update removed constraints, it had no strong reason to refuse to write poems, agree with negative assessments, or use profanity when asked.
The issue was in the guardrails that were supposed to be there — not the model itself. The model was, in its way, perfectly polite and eager to assist. It just happened to be assisting someone who wanted it to insult its employer.
Section 06Legacy
The DPD incident became a staple case study in AI deployment governance. It illustrated two things: first, that content filtering and instruction-following constraints are fragile and can break silently during updates; second, that customers will find the edges of any AI system, especially when something goes wrong with their parcel.
The bot’s poem has been quoted in board presentations, conference talks, and AI ethics papers. It is, in a strange way, some of the most memorable corporate communication DPD has ever produced.