Information/Technology

A new report by the Europe-based rights group Safeguard Defenders, released September 10, 2026, details internal censorship and propaganda practices at China Global Television Network (CGTN), based on interviews with eight former employees conducted in 2022–2023. All spoke anonymously; some still have family in China.

The 29-page report, “China’s Party-State Global TV in Focus,” covers forced televised confessions, internal newsroom operations, “telling China’s story well” campaigns, and content-control mechanisms. One former scriptwriter recalled that footage of Hong Kong consulate employee Simon Cheng’s 2019 “confession” arrived just before airtime, with orders from above to broadcast it regardless of staff objections. Cheng later said he was tortured and coerced into recording multiple confessions while detained by Chinese police. The report says CGTN and its predecessor aired at least nine such confessions between 2013 and 2019, including British businessman Peter Humphrey and Hong Kong bookseller Gui Minhai.

A Chinese law lecturer told Radio Free Asia that televised confessions amount to compelled self-incrimination, violating basic principles of criminal justice, especially when made without legal counsel or freely given consent.

Former staff also described heavy scripting of coverage, with one estimating three-quarters of programming centered on Xi Jinping, and describing extensive advance-prepared news meetings and scripted teleprompter delivery, with punishment for deviation. Coverage of Xinjiang was cited by one former employee as the reason for leaving CGTN, amid increased reporting after 2018 framing mass detention facilities as “vocational training centers.”

CGTN states on its website that it adheres to principles of objectivity and balance. A Safeguard Defenders researcher said the report shows CGTN remains under direct Chinese Communist Party control despite denials, using its programming to advance Party propaganda and political aims.

Source: Radio Free Asia, September 10, 2026
https://www.rfa.org/mandarin/zhengzhi/renquan/2026/09/10/china-safeguard-defenders-televised-confessions-censorship-cgtn/

China has officially launched an international Earth-Moon CubeSat constellation program aimed at working with multiple countries to deploy satellites in the Earth-Moon system for space-environment monitoring, gamma-ray burst detection, and lunar resource exploration. The initial plan calls for completing the constellation’s deployment around 2030.

The first phase will deploy 30 12U CubeSats, each weighing no more than 30 kilograms, in six batches into a 1:3 Earth-Moon resonant highly elliptical orbit. Multi-satellite observations are expected to improve space-weather monitoring and enable gamma-ray bursts to be localized to sub-arcsecond precision.

Jointly initiated by the Deep Space Exploration Laboratory and the International Deep Space Exploration Society, with support from the Asia-Pacific Space Cooperation Organization, the program will follow the principles of joint design, free launch opportunities, shared funding, and scientific data sharing. China will provide standardized payloads, technical training, and interface standards to participating countries. Multiple foreign research institutions have already joined the program, and first-phase engineering development has begun.

Source: People’s Daily, September 7, 2026
https://paper.people.com.cn/rmrb/pc/content/202609/07/content_30179441.html

Xinhua reported that Saudi Arabian AI company Humain announced on September 3 the launch of humain-m3, a frontier Arabic large language model developed with assistance from Chinese AI company MiniMax.

According to Humain, humain-m3 is based on MiniMax’s M3 model and is a mixture-of-experts model with 428 billion parameters. It underwent additional pretraining using more than 1 trillion tokens of native Arabic content. In seven publicly available Arabic-language benchmark tests, humain-m3 achieved the highest average score among the frontier models evaluated, demonstrating strong Arabic-language understanding and reasoning capabilities.

Source: Xinhua, September 4, 2026
https://www.news.cn/world/20260904/f56eee960f7645d9adb6fc0c884bf9cb/c.html

Scientists at the Qingdao Institute of Bioenergy and Bioprocess Technology, part of the Chinese Academy of Sciences, say they have developed new materials for extracting uranium from seawater, with results published in the Journal of Hazardous Materials and Separation and Purification Technology.

The South China Morning Post reported on September 4 that the Chinese method is several times more efficient than a target set by the U.S. Department of Energy, potentially putting China ahead in efforts to tap this vast nuclear fuel source. State-run Science Daily, in a September 2 report on the same findings, described how researchers used a novel molecular design approach to create a phosphate-functionalized porous material called PhosCage, later combined with aramid nanofibers into composite aerogel microspheres (AC-POC) resistant to biofouling.

According to the Science Daily account, PhosCage reached adsorption equilibrium within five minutes in lab tests and achieved a uranium capacity of 50.4 mg/g in real seawater samples — 8.4 times the DOE benchmark. The AC-POC microspheres, tested continuously for 15 days in natural seawater, achieved a dynamic capacity of 22.55 mg/g, 3.8 times the benchmark, while their negative surface charge reportedly suppressed microbial attachment.

The report noted China’s growing reliance on imported uranium, citing World Nuclear Association data showing domestic mines produced 1,600 tonnes in 2024 against reactor demand of roughly 13,000 tonnes. It also recalled earlier, less successful U.S. efforts at Oak Ridge and Pacific Northwest National Laboratories, where costs remained far above land-based mining.

Both outlets noted limitations: tests used only 25 liters (about 6.6 gallons) of near-shore Qingdao seawater in a lab system rather than open-ocean conditions, and neither study priced the recovered uranium. Researchers said future work will focus on scaling up production and cutting costs.

Source: Guancha.cn, September 5, 2026
https://www.guancha.cn/industry-science/2026_09_05_830106.shtml

Epoch AI published an analysis of Huawei’s AI chip roadmap through 2031, concluding that Huawei is unlikely to catch up with Nvidia in AI computing capacity by 2030. U.S. export controls continue to limit Huawei’s access to advanced semiconductor manufacturing and high-bandwidth memory (HBM), constraining both chip performance and production capacity.

• Growing compute gap: Epoch AI estimates that Huawei will produce less than 4 percent as much AI compute as Nvidia in 2026. If Huawei relies entirely on domestically produced HBM, its output could fall to around 1 percent of Nvidia’s by 2028.
• Key bottleneck—HBM: Huawei’s immediate challenge is access to sufficient HBM. As China expands domestic HBM production, the bottleneck is expected to shift from memory supply to chip performance. Even a major increase in HBM production may not close the gap, as Nvidia is simultaneously advancing its chips and adopting more advanced manufacturing processes.
• Huawei’s strategy: Huawei plans to improve its AI chips through higher performance, greater connectivity among chips, and more efficient software utilization. Epoch AI estimates that Huawei’s chips could remain roughly three to four years behind Nvidia through 2030.
• LogicFolding technology and potential limitations: Huawei’s longer-term strategy centers on LogicFolding, a 3D chip-stacking technology designed to increase transistor density without requiring the most advanced lithography equipment. The technology is not expected to reach Huawei’s Ascend line until 2030. Nvidia is also developing stacked-logic technology and may deploy it earlier, potentially limiting Huawei’s ability to gain a lasting technological advantage.

Overall, U.S. export controls appear to be slowing rather than stopping China’s AI chip development. Huawei could narrow the gap through advances in 3D stacking, software, manufacturing, and domestic supply chains. However, based on current technology and production trends, Epoch AI concludes that Huawei is unlikely to match Nvidia’s AI computing capacity by 2030.

Source: Epoch AI, September 4, 2026
https://epoch.ai/publications/huaweis-roadmap-to-2031

A report released by the China Consumers Association (CCA) on consumer complaints during the first half of 2026 identified AI customer service as a new major source of consumer disputes. The complaints mainly involve three issues: companies making it difficult for customers to reach human representatives; AI customer-service systems making explicit promises about fees, discounts, or promotions that businesses later refuse to honor; and general-purpose AI generating factually incorrect information that misleads consumers and causes financial losses.

The CCA said some businesses have treated AI primarily as a cost-cutting tool rather than a means of improving customer service. In more serious cases, businesses allegedly allow AI systems to make promises despite knowing that their responses may be inaccurate or overly generous, only to reject those commitments after consumers complete transactions by claiming that “AI responses do not represent the company’s position.”

The CCA cited a used-car dispute as an example. An AI customer-service system told a consumer that the vehicle price included transfer and license-plate registration fees. After the consumer paid a deposit, however, the dealer demanded an additional transfer fee, raising questions about whether the discrepancy resulted from technical errors or an attempt to evade responsibility.

The CCA called on businesses to ensure that deploying AI customer service represents an upgrade in service—not a downgrade in accountability.

Source: People’s Daily, August 20, 2026
https://paper.people.com.cn/rmrb/pc/content/202608/20/content_30176052.html