AGCM extracts binding commitments from DeepSeek, Mistral and Nova AI over AI hallucinations
Italy's AGCM extracted binding commitments from AI firms DeepSeek, Mistral and Nova AI regarding AI hallucinations after probes; the case closed with these commitments in place and no infringement findings.
Chatbots failed to clearly warn users that AI-generated outputs could be fabricated.
Key facts
- What
- Italy's AGCM extracted binding commitments from AI firms DeepSeek, Mistral and Nova AI regarding AI hallucinations after probes; the case closed with these commitments in place and no infringement findings.
- Incident date
- Apr 30, 2026
- Who
- DeepSeek, Mistral, Nova AI
- Failure mode
- Hallucination
- AI surface
- Chatbot
- Severity
- Medium
What happened
Italy's AGCM concluded probes into DeepSeek, Mistral, and Nova AI after finding their chatbots lacked sufficient warnings about potential hallucinations. The companies agreed to binding commitments to provide clear warnings about hallucination risks, and the proceedings were closed.
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.
The system did not explicitly warn that outputs could be fabricated, reducing transparency about factual reliability. This disclosure gap contributed to user misinterpretation of chatbot outputs.
What it cost
Sources
Cite this entry
https://failureindex.ai/failures/agcm-extracts-binding-commitments-deepseek-mistralAI Failure Index. "AGCM extracts binding commitments from DeepSeek, Mistral and Nova AI over AI hallucinations" (FI-0320). Realm Labs. https://failureindex.ai/failures/agcm-extracts-binding-commitments-deepseek-mistral (indexed Jun 8, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0320. 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.