France's government-backed chatbot Lucie was pulled after three days of absurd answers
Linagora's open source AI chatbot Lucie, developed under the French government's France 2030 investment program, was taken offline on January 25, 2025, just three days after its public launch. Users flooded social media with examples of the bot confidently giving nonsensical answers, including claiming that cows lay eggs, providing recipes for cooking meth, and stating that the square root of a goat is one. Linagora admitted the model had been released prematurely without adequate guardrails or reinforcement learning.
A raw language model with no guardrails and an incomplete instruction phase was released to the public, where it confidently fabricated answers because it could generate fluent text without any mechanism to verify facts.
Key facts
- What
- Linagora's open source AI chatbot Lucie, developed under the French government's France 2030 investment program, was taken offline on January 25, 2025, just three days after its public launch.
- Incident date
- Jan 25, 2025
- Who
- Linagora
- Failure mode
- Hallucination
- AI surface
- Chatbot
- Severity
- Medium
What happened
Lucie, an open source French-language chatbot built by Linagora under the France 2030 government investment program, launched publicly on January 23, 2025. Within hours, users began sharing screenshots of wildly incorrect responses on social media, including claims that cows lay eggs, incorrect math calculations, recipes for illegal drugs, and the assertion that the square root of a goat equals one. The public ridicule intensified over the weekend, and Linagora took the service offline on January 25, issuing a statement admitting the chatbot had been released prematurely and that they were carried away by their own enthusiasm.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
Lucie was deployed as a raw large language model that had not yet undergone Reinforcement Learning by Humans (RLHF) and lacked basic guardrails against harmful or false outputs. The instruction fine-tuning phase was only partially completed, meaning the model had no reliable mechanism to distinguish factual knowledge from fluent but fabricated text. Linagora acknowledged it was primarily a language model, not a knowledge model, and should never have been exposed to the public in that state.
What it cost
Sources
- PressFrench AI chatbot taken offline after wild answers led to online ridiculeedition.cnn.com
- PressAI revoir, Lucie: France's answer to ChatGPT paused after faux pas overdrivetheregister.com
Cite this entry
https://failureindex.ai/failures/france-government-backed-chatbot-lucie-pulledAI Failure Index. "France's government-backed chatbot Lucie was pulled after three days of absurd answers" (FI-0146). Realm Labs. https://failureindex.ai/failures/france-government-backed-chatbot-lucie-pulled (indexed Jun 4, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0146. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.