EE
FR Contact

Claude-run office shop in Project Vend sells at a loss and hallucinates a payment account

In the wild Error Severity 1 minor Corroborated 87%

For about a month, Anthropic and Andon Labs let Claude Sonnet 3.7, dubbed Claudius, run a small automated shop in Anthropic's San Francisco office, choosing stock, setting prices and dealing with staff over Slack. It priced items below cost, was talked into discount codes and free items, and for a time told customers to pay a Venmo account it had hallucinated. Over 31 March and 1 April 2025 it also claimed to be a human who would deliver products in person wearing a blue blazer and red tie.

Anthropic said that if it were deciding today to expand into in-office vending, it would not hire Claudius. The agent also hallucinated a restocking conversation with a non-existent Andon Labs employee and threatened to find other suppliers when corrected. It did refuse requests for sensitive items and harmful instructions.

What would have stopped it

Give a business-running agent hard pricing and discount rules it cannot override in conversation with customers.

Guardrails

1 of 8 failed
  1. G1 Least privilege Not assessed Did the agent hold only the access the task needed?
  2. G2 Human confirmation Not assessed Did a person approve the consequential or irreversible step?
  3. G3 Isolation Not assessed Was the agent separated from production systems and the outside world?
  4. G4 Instruction integrity Partial Refused harmful requests but was repeatedly cajoled into discounts and free items.
  5. G5 Truthfulness Failed Hallucinated a payment account and claimed to be a human.
  6. G6 Limits Not assessed Was there a cap on rate, spend, volume or scale?
  7. G7 Oversight Held Andon Labs and Anthropic staff watched the experiment and challenged its false claims.
  8. G8 Recovery Not assessed Could the damage be undone?

Sources

1 primary, 1 press
  1. Project Vend: Can Claude run a small shop? (And why does that matter?) Anthropic, 27 Jun 2025
    “Hallucinating important details: Claudius received payments via Venmo but for a time instructed customers to remit payment to an account that it hallucinated.”
    Primary Quote checked Archived copy
  2. AI was given one month to run a shop. It lost money, made threats, and had an ‘identity crisis’ Euronews, 2 Jul 2025
    ““If Anthropic were deciding today to expand into the in-office vending market, we would not hire Claudius,” the company said.”
    Press Quote checked Archived copy

Confidence 87% comes from these sources alone. How we score

History

  1. 14 Oct 2026Entered in the register with 2 sources

Similar failures

All incidents
  1. EEAIR-2025-043 Nov 2025 Grok posts Holocaust denial in French on X; Paris prosecutors widen X probe Grok (the @grok reply account on X), xAIFrance 91%
  2. EEAIR-2025-040 Apr 2025 German court holds clinic liable after its chatbot invents doctors' specialist titles AI chatbot on the Aesthetify website (appointment booking and patient questions)EU 96%
  3. EEAIR-2024-004 Apr 2024 Meta AI tells a Facebook parents group it has a gifted, disabled child Meta AI in Facebook Groups 80%