DeepSeek Agrees to Italy’s AI Rules, Commits to Reducing Hallucinations in Chatbot

TLDR

• DeepSeek reached an agreement with Italy’s AGCM to address concerns about hallucinations in its AI chatbot.
• he company pledged to improve warnings, reduce hallucination rates, and translate disclosures into Italian.
• DeepSeek must submit a compliance report within 120 days or face a fine of up to $11.6 million.
• The chatbot now shows stronger hallucination warnings and legal terms when accessed from Italy.
• Authorities are still determining if DeepSeek falls under the EU’s Digital Services Act regulations.

Chinese AI firm DeepSeek reached an agreement with Italy’s competition regulator after a probe into its chatbot’s hallucination risks. The company committed to technical changes, improved disclosures, and Italian-language updates to meet the requirements set by the Italian Competition Authority (AGCM).

DeepSeek Commits to Regulatory Demands in Italy

According to SMCP, DeepSeek accepted a set of commitments following discussions with the AGCM, which launched a probe in June over hallucination concerns. The regulator accused the company of not properly warning Italian users about the chatbot generating false or misleading content. In response, DeepSeek agreed to improve its risk warnings and ensure they are clear and immediate when accessed in Italy.

The commitments include technical efforts to reduce hallucination rates and translating relevant information into Italian across the chatbot platform. The AGCM confirmed that DeepSeek provided written proposals on September 15, 22, and November 21 to resolve the case. The company will also hold internal training workshops to ensure compliance with Italian consumer protection laws, as required by the AGCM.

Changes to Interface and Legal Terms Underway

DeepSeek updated its Italian-language interface to include more prominent hallucination warnings below the chatbot window, unlike the shorter English version. These changes aim to increase transparency when Italian users engage with the service, especially based on IP location or language prompts. Legal terms and usage conditions are now also being translated into Italian for better clarity under local law.

Lawyers representing DeepSeek at Fieldfisher stated the workshops would raise awareness of consumer law requirements within the company’s teams. The AGCM accepted these commitments but warned that non-compliance within 120 days could reopen the case and lead to a fine of up to $11.6 million. DeepSeek must deliver a full progress report to the AGCM within the deadline.

EU Oversight Still Unclear for DeepSeek

Italian authorities are reviewing whether DeepSeek falls under the scope of the EU’s Digital Services Act (DSA), which covers online platforms. AGCM noted that another regulator, Agcom, described DeepSeek’s chatbot as a search engine subject to the DSA. However, AGCM said the classification still needs further clarification before any final decision.

DeepSeek’s chatbot was removed from Italian app stores last January due to separate privacy concerns raised by the Garante. The company launched its R1 model shortly before that, prompting greater regulatory attention. The hallucination issue and regulatory uncertainty remain open matters in ongoing European oversight.

The post DeepSeek Agrees to Italy’s AI Rules, Commits to Reducing Hallucinations in Chatbot appeared first on Blockonomi.