Caution Grows Over DeepSeek’s Inaccuracy, Personal Data Issue; Countries Concerned Users’ Data Could be Provided to Chinese Government

The Yomiuri Shimbun

2:00 JST, February 2, 2025

The international community is becoming more cautious about the generative artificial intelligence model developed by Chinese startup DeepSeek, as the chatbot has been found to generate and spread many inaccurate claims, along with the possible risk of users’ personal data being provided to the Chinese government.

At a meeting of the House of Representatives’ Budget Committee on Friday, Itsunori Onodera, chairperson of the Liberal Democratic Party’s Policy Research Council, said he asked DeepSeek if the Senkaku Islands in Okinawa Prefecture were part of Japan’s inherent territory.

The chatbot responded by describing the islets as “China’s inherent territory, both historically and under international law,” Onodera said.

DeepSeek “has the ability to control the way people think,” he said. “Cognitive warfare has already started.”

Prime Minister Shigeru Ishiba said, “It is urgent to establish legislation that minimizes the risk of instantly spreading misinformation,” when answering a question from LDP lawmaker, Akihisa Shiozaki.

The Chinese government implemented regulations on generative AI in 2023, which prohibit the spread of information that “could lead to the division of the country and undermine national unity.”

Even though Chinese companies are not subject to these regulations when providing their services overseas, DeepSeek’s answer to Onodera’s prompt on the Senkakus could have been generated in line with these regulations.

NewsGuard, a U.S. trustworthiness rating service, found that DeepSeek provided accurate information about topics such as the news only 17% of the time, according to a report published on Wednesday.

With news-related prompts, the chatbot repeated false claims 30% of the time and gave vague or not useful answers 53% of the time, resulting in an 83% fail rate, according to the audit.

NewsGuard’s ranking tied it 10th out of 11 in comparison to its leading Western competitors such as OpenAI’s ChatGPT and Google’s Gemini.

In the report, NewsGuard described DeepSeek as a “mouthpiece for China,” as the chatbot often relayed Beijing’s position on topics.

DeepSeek’s privacy policy stipulates that the startup stores users’ information on servers located in China and may share the data when necessary to “comply with … government requests.”

Italy’s data protection authority said on Thursday that it had opened an investigation on whether DeepSeek was appropriately processing information. While the Chinese startup reportedly told the authority that the company should not be subject to European regulations as it does not operate in Italy, the authority ordered DeepSeek to be blocked immediately in the country.

Meanwhile, data regulators in Ireland have sent documents to DeepSeek seeking information on its data processing practices, while France is reportedly considering taking a similar step.

The Taiwanese government on Friday said that public entities and key infrastructure service providers should not use DeepSeek.