Chinese artificial intelligence firm DeepSeek has agreed to roll out the first country-specific version of its chatbot and to curb so-called “hallucinations” in its models to meet regulatory requirements in Italy.

The Hangzhou-based start-up reached the agreement after months of negotiations with Italy’s Competition Authority, the AGCM, which had accused the company of failing to adequately warn users about the risk that its chatbot could generate incorrect or misleading information.

Under the commitments, DeepSeek will introduce an Italy-specific version of its namesake chatbot and adopt measures aimed at reducing hallucinations and improving user disclosures, according to people familiar with the matter.