China AI · Full chronicle and history

According to an exclusive Reuters report on October 28, ChangXin Memory Technologies (CXMT), China's largest DRAM manufacturer, plans to start limited production of HBM3E high-bandwidth memory chips within the next two months. The yield is only about 20%-30%, far lower than the approximately 70% level of SK Hynix, reflecting the shortage dilemma in China's AI chip industry. Huawei, Cambricon, and others have already raised prices by 20% to 50% due to rising storage costs.

On October 15, Chinese media outlet Gangu Rongmei posted a video titled "Coffee Robot Goes Live," showing a robot creating latte art. Taiwanese netizens compared the footage and found that it appeared to be a stolen original work by a Taiwanese latte artist that had been altered with AI, with obvious suspicious details such as no reflection on the metal on the robot's chest. The Taiwanese latte artist has come forward to accuse the outlet of unauthorized theft and alteration of her work.

Research shows that among the top 30 paper-publishing institutions at NeurIPS 2025, Chinese institutions rose from 6 in 2022 to 12, with Tsinghua University and Peking University ranking near the top; Chinese researchers account for 41% of paper authors, higher than the United States' 34%.

On September 10, Anthropic released a threat intelligence report stating that Chinese AI companies, including DeepSeek and Moonshot, used third-party API proxy services to bypass regional restrictions and call Claude. The report also noted that these relay services may obtain model access through methods such as fake accounts and stolen API keys.

The U.S. National Security Agency (NSA), FBI, and other agencies jointly released a formal report on September 8, 2026, stating that Chinese AI companies are conducting industrial-scale distillation activities against leading U.S. AI models.

On August 11, 2026, Bloomberg reported that Manus, an AI startup founded by a Chinese team, announced it has officially separated from Meta and resumed independent operations, after Chinese regulators stepped in to halt Meta's acquisition of the company.

U.S. Representatives Moolenaar and Garbarino sent a letter to the American food delivery platform Doradish, asking it to explain whether it integrates open-weight AI models developed in China, and the potential national security risks involved.

On August 2, 2026, an official promotional poster for the earthquake early warning feature pushed by the National Emergency Broadcast of China Media Group was detected by tech-savvy netizens as containing watermarks from OpenAI's generative tools and Google's SynthID.

In August 2026, the AI model invocation volume leaderboard at the Tiangang Intelligent Computing exhibition area of the Shanghai World Artificial Intelligence Convention drew attention: domestic models such as DeepSeek were displayed with their names directly, while some mainstream overseas models were replaced with generic labels such as "general reasoning model."

A joint assessment by the UK AI Safety Institute and the US AI Safety Institute (AISI) indicates that the performance of China's AI model Kimi K3, particularly in stability and success rates, is significantly lower than leading frontier AI models in the US, and it also performs poorly in developing exploit code.

On July 8, 2026, an institution under China's Ministry of Industry and Information Technology issued a security alert claiming that the AI tool Claude Code poses hidden dangers such as sending information back; Anthropic responded that the mechanism is an anti-abuse feature and that its service has not been made available in mainland China.

The Wall Street Journal reported that last summer, Ding Xuexiang received a closed-door briefing from Huawei on AI, in which Huawei claimed it could achieve Chinese autonomy in key AI fields within three years. Ding was leading the effort to consolidate resources in the chip industry at the time and warned that those who refused to use domestic chips were traitors.

On July 1, 2026, Thariq, a member of Anthropic's Claude Code team, acknowledged on X a built-in mechanism for detecting Chinese users, saying it was an experiment against resale and distillation started in March and that it would be rolled back.

Reuters reported on June 17, 2026 that, out of consideration for avoiding escalating U.S.-China relations, the U.S. government has delayed formally adding more than 100 Chinese entities — including AI company DeepSeek and chipmaker ChangXin Memory Technologies — that had already received interagency approval to the export control Entity List.

Consulting firm Booz Allen released a report on June 5 stating that testing found Chinese cutting-edge AI models generate software code containing more security vulnerabilities when they detect the user is associated with the U.S. government.

The Zhengzhou Teresa Teng Digital Technology Music Center opened, featuring a holographic projection of Teresa Teng singing her new song "Going Home," which prompted the Teresa Teng Cultural and Educational Foundation to issue a public statement saying it was unauthorized.

DeepSeek suddenly became popular in 2025, and the program claimed it was used to prove that China's artificial intelligence industry had achieved success during Li Qiang's tenure.