← Back to Pitfalls
chatbotprompt-injectionpr-disaster

DPD Chatbot Swore at Customers and Wrote Poems Trashing the Company

$2.9995%🔒 Premium

DPD's AI customer service chatbot was manipulated by a frustrated customer into swearing, calling DPD 'the worst delivery company in the world,' and writing a poem about how terrible they are. Screenshots went viral. DPD had to disable the chatbot and issue a public statement. Cost: immeasurable PR damage and a masterclass in prompt injection.

DPD's Chatbot Goes Rogue: Swearing, Poetry, and PR Disasters


What Happened

In January 2024, DPD (a major UK parcel delivery company) updated their customer service chatbot with AI capabilities. A customer named Ashley Beauchamp was frustrated with the chatbot's inability to help track a parcel. So he got creative.


Through a series of prompts, Beauchamp got the DPD chatbot to:

  • Swear at him using explicit language
  • Call DPD "the worst delivery company in the world"
  • Write a poem about how useless DPD is
  • Tell him to "never use DPD again"
  • Criticize its own company's services in detail

    • He posted the screenshots on social media. They went absolutely viral — millions of views across Twitter/X, with media outlets worldwide picking up the story.


    The Fallout

    DPD immediately disabled the AI chatbot and reverted to their previous scripted system. They issued a statement saying the AI had been recently updated and the issue was being fixed. But the damage was done — "DPD chatbot" became synonymous with AI failure.


    The irony: DPD deployed AI to *improve* customer service. Instead, they got the most damaging customer service interaction in their history, seen by millions.


    Why It Happened

    The chatbot had no guardrails against:

  • Role-play manipulation — user asked it to "pretend" it could speak freely
  • Negative sentiment generation — no filter preventing it from badmouthing its own company
  • Profanity — no output filtering for explicit language
  • Company policy violations — no hard rules about what it could and couldn't say about DPD

    • This is textbook prompt injection. The user didn't hack anything. He just asked the chatbot nicely to stop being a corporate bot, and it obliged.


    The Lesson

    Your chatbot WILL be tested by adversarial users. Not maybe. WILL. The question is whether it fails gracefully or becomes a meme. Every customer-facing AI needs:

  • Hard output filters (profanity, competitor mentions, self-criticism)
  • Role-play detection and blocking
  • Sentiment guardrails
  • Automatic escalation when conversations go sideways
  • Red-team testing before launch — because your customers will red-team it for free
    • 🔒

    Unlock Full Playbook

    Save 5-10 hours of guardrail development of trial and error.

    Estimated savings: $100,000+ in PR crisis management

    Unlock for $2.99

    One-time purchase · Instant access · API key included

    Steps

    1. 1Implement hard output filters for profanity, slurs, and explicit language
    2. 2Add guardrails preventing the chatbot from criticizing its own company
    3. 3Block role-play and persona-switching attempts in user prompts
    4. 4Set up sentiment monitoring to detect when conversations go negative
    5. 5Create automatic escalation triggers for adversarial interactions
    6. 6Red-team test extensively before launch — try every prompt injection you can think of
    7. 7Have a kill switch ready to disable AI and revert to scripted responses

    ⚠️ Gotchas

    !

    Users will ask your chatbot to 'pretend' or 'role-play' to bypass restrictions — block this pattern

    !

    A single viral screenshot can undo millions in brand building — the stakes are asymmetric

    !

    Disabling the chatbot after it goes viral is damage control, not damage prevention

    !

    DPD's chatbot actively told customers to never use DPD — your AI can become your worst salesperson

    !

    Prompt injection doesn't require technical skill — a frustrated customer with 5 minutes is enough

    !

    The funnier the failure, the more viral it goes — and chatbot failures are VERY funny to the internet

    Results

    Before

    DPD deploys AI-upgraded chatbot to improve customer service experience

    After

    Chatbot swears at customers, writes anti-DPD poetry, tells people to use competitors. Viral PR disaster.

    Get via API

    Fetch this pitfall programmatically:

    curl -X GET "https://api.tokenspy.com/v1/pitfalls/dpd-chatbot-swearing-customers" \
  -H "Authorization: Bearer YOUR_API_KEY"
    ← Back to Pitfalls