Welcome to the HRIC Digital Rights Report, our month-to-month roundup of developments in China’s digital rights panorama, together with cybersecurity, AI, digital transnational repression, censorship, the Nice Firewall, and web freedom.
Devastating flash floods alongside the Nepal–Tibet border in late August uncovered the human value of China’s info controls. Whereas greater than 1,000 folks have been confirmed lifeless and 1000’s remained lacking throughout the affected area, Chinese language authorities minimized details about casualties, restricted unbiased reporting, and detained folks for sharing details about the catastrophe.
Eyewitness footage disappeared from Chinese language social media, whereas authorities used image-recognition know-how to determine and take away delicate content material and flooded on-line areas with official materials to bury unbiased accounts. Authorities reportedly detained or fined a minimum of 10 folks for questioning official accounts of the catastrophe on-line.
This response mirrored a broader sample in China of prioritizing “social stability” over the free circulate of data. Beneath this strategy, authorities limit info that might generate public criticism, expose official failures, or result in calls for for accountability. On the similar time, state media promote fastidiously managed narratives that emphasize social order and the federal government’s capacity to reply to crises. Following the Nepal–Tibet floods, for instance, Chinese language state media targeted on rescue operations and the federal government’s response whereas downplaying the complete scale of the destruction. Equally, 5 years after the lethal 2021 Zhengzhou floods, in August, authorities continued censoring posts that commemorated victims and preserved public recollections of the catastrophe.
This strategy additionally prolonged past disasters and explicitly political speech. Authorities eliminated a simple WeChat put up describing the exhausting safety checks required to go to Beijing’s Zhengyangmen gate, illustrating how even bizarre frustration may be censored when it displays negatively on the federal government’s safety agenda. Comparable restrictions affected expressions of grief in Hong Kong, the place organizers encountered rejected commercials and shuttered memorial shows after lots of gathered to mourn Wang Wang, a stray canine tortured and killed in Guangdong.
Though such censorship could defend the federal government from criticism and protect an look of stability within the quick time period, it comes on the expense of public security and entry to dependable info. Throughout a catastrophe, well timed reporting may help residents perceive the dangers they face, find lacking family members, determine areas in want of help, and make knowledgeable selections about whether or not to evacuate. Proscribing that info can delay emergency responses, gas rumors, weaken public belief, and make it harder to find out the true scale of the harm.
The consequences of those controls will not be restricted to emergencies. A WeChat essay translated by China Digital Instances this month described China’s keyword-based censorship as blunt and counterproductive, arguing that it blocks innocent speech, fails to deal with real harms, and forces customers to speak via riddles. The essay highlights how broad and automatic restrictions can intervene with bizarre communication, even when customers will not be criticizing the federal government or discussing politically delicate points.
The examples coated this month—from catastrophe footage and complaints about safety checks to memorials for a stray canine—completely illustrate this downside. They present that speech doesn’t have to resemble typical political dissent to change into a goal in China. Censorship more and more shapes not solely what folks could criticize, but additionally what they could know, witness, mourn, and say about their on a regular basis lives. Because the boundaries of restricted speech proceed to increase, the excellence between controlling political dissent and controlling bizarre public expression turns into more and more tough to attract.
Translation: “You Can’t Deal with Sexual Assault at Banquets With out Addressing the Nature of Energy”: A censored WeChat put up examined two latest instances of alleged sexual assault by Chinese language officers, arguing that piecemeal responses to particular person scandals fail to deal with the systemic difficulty: the shortage of institutional checks on official energy that permits such abuse to recur.
Translations: “Typically Forgetting Is as Horrible because the Catastrophe.” Censored Reflections on the Lethal Zhengzhou Floods, 5 Years On: Marking 5 years since Zhengzhou’s lethal 2021 floods (398 lifeless, many deaths hid), two censored Chinese language essays mirror on reminiscence, grief, and the “Raincoat Dad” who mourned his drowned daughter, warning that societal amnesia permits man-made disasters to recur. Posts merely commemorating the tragedy, with out demanding accountability, have been nonetheless censored.
Translation: No Scaling the Nice Firewall within the Library: A Hunan Library discover suspended library Wi-Fi after patrons have been caught utilizing censorship circumvention instruments, publishing partly redacted telephone numbers of offenders as a warning that web use is tied to actual identities. Commenters famous circumventing on private cell knowledge is safer than on institutional networks, which face heavier scrutiny and retaliation towards directors.
HRIC on Twitter/X: There was a latest crackdown on content material referencing a standard bamboo “cicada” toy as satirical commentary on Huawei. The toy, which emits a steady “wah-wah” sound when shaken, was observed by netizens for its sound impact just like the synchronized, exaggerated gasps of shock from the viewers at Huawei launch occasions. Searches for “bamboo cicada” on Weibo have been cleaned, whereas Douyin, Kuaishou, Xiaohongshu, and WeChat public accounts noticed a flood of account restriction notices. On August 2, “Is it overkill to take down Yu Chengdong parody movies?” trended on scorching searches, with some state-affiliated media beginning to defend the removals and bans.
Hanging a Banner for Chinese language AI: Folks’s Day by day just lately ran a symmetrical, parallel-column entrance web page proclaiming Xi Jinping’s “concern” for each home AI growth (”useful, secure, equitable”) and China’s function in bridging the worldwide AI divide for growing nations. CMP explains: “The bizarre column design, with its fastidiously conceived symmetry—a propaganda appropriation of the normal parallel saying, or duilian (对联)—is supposed little question to be a strong parallelism asserting the Chinese language Communist Celebration management’s twin convictions on synthetic intelligence, for the nation and for the world.”
HRIC on Twitter/X: The Wall Road Journal reported that a number of new research have discovered U.S. synthetic intelligence fashions unconsciously exhibit similarities to China’s censored chatbots when answering sure delicate questions, whereas AI firms have made restricted efforts to deal with this difficulty.
The Downside with Peach: Chinese language AI-generated influencer “Peach Fang,” star of a viral AI drama with 200 million views, sparked controversy after selling contact lenses she can not bodily put on or check, elevating questions on AI personas making false claims to bodily expertise. State media and attorneys famous AI “endorsers” lack authorized personhood, exposing regulatory blind spots round artificial advertising and marketing.
Translation: “AI Has Decided That My Press Card Is a Faux”: A China Youth Day by day journalist recounts the surreal expertise of an interview topic’s AI chatbot confidently declaring his official press card a pretend, illustrating the rising, typically absurd, methods AI instruments are reshaping belief, misinformation, and censorship dynamics in Chinese language journalism and society.
HRIC on Twitter/X: On August 5, U.S. cybersecurity firm VulnCheck launched a analysis report stating that routers shipped from Shenzhen Zhibotong Electronics (Zbtlink/ZBT) come pre-installed with a backdoor program of their firmware. As quickly because the gadget is powered on, it constantly initiates outbound connections to a number of mounted abroad command-and-control servers, with two main nodes hosted on Alibaba Cloud in Shanghai and Shenzhen, respectively.
Royal Navy drone cameras linked to China increase safety issues: Cameras on the UK Royal Navy’s Chinese language-component K3 Scout surveillance drones, utilized by UK particular forces, have been discovered sending routine “heartbeat” alerts to a tool in China throughout a cybersecurity evaluation. The UK Ministry of Defence acknowledged that no delicate knowledge was compromised however has reduce the drones’ web connectivity. The incident provides to mounting concern over Chinese language-made elements embedded in Western navy and important infrastructure programs.
Hong Kong mulls AI facial recognition-enabled drones to police smoking ban at building websites, 39 fines issued thus far: Hong Kong’s Labour Division has issued 39 fines and 11 enchancment notices since a construction-site smoking ban took impact in mid-July, and is now exploring including AI facial recognition to its infrared drones (already used to detect cigarette warmth signatures) to raised determine offenders, with growth anticipated to take 9–12 months. The transfer raises questions concerning the expanded use of AI facial recognition drones for surveillance functions.
Hong Kong police arrest 10 for harmful driving in first drone-assisted site visitors crackdown: Hong Kong police arrested 10 drivers for harmful driving on the Kwun Tong Bypass of their first drone-assisted site visitors operation, utilizing aerial surveillance alongside unmarked “stealth vehicles” to collect proof.
HRIC on Twitter/X: A New York Instances’ investigation right into a latest leak from cybersecurity researcher Mark Hofner on the “International Personnel Dynamic Management Platform” reveals new particulars concerning the software, which tracked over 700 international nationals residing regionally. Greg Walton, a senior investigator on the Canadian SecDev Group, described the incident as a symptom of “surveillance proliferation” pushed by an ever-expanding ecosystem of distributors, contractors, and public safety departments: every extra platform creates extra locations the place delicate private knowledge could possibly be misconfigured, copied, or uncovered.
From petitions to memes: How college students made dissent seen: The piece traces how Chinese language scholar dissent has advanced from formal petitions towards subtler, coded varieties like memes, puns, and symbols (echoing ways from the 2022 “White Paper Motion”) to evade censorship whereas nonetheless expressing discontent. This shift displays each know-how’s function in activism and authorities’ tightening crackdown on any direct problem to the state.
Nvidia senior supervisor linked to Supermicro scheme smuggling AI servers to China: Taiwan indicted 9 folks, together with a senior Nvidia supervisor and two Supermicro workers, for allegedly forging paperwork to smuggle 74 AI servers with restricted B300 chips to China by way of Japan, Indonesia, and Hong Kong, evading US export controls.
The rise of AI climate forecasting as China emerges as a pacesetter in race to enhance predictions: As Storm Dolphin approached, Chinese language AI climate fashions like Shanghai AI Lab’s Fengwu, Huawei’s Pangu, and Fudan’s Fuxi labored alongside conventional forecasting, with Fengwu predicting the hurricane’s landfall time and site inside half-hour and 30km 5 days out. China is rising as a pacesetter in AI climate prediction, although such fashions nonetheless lag conventional programs in forecasting storm depth.
ByteDance trains large AI mannequin in bid to rival Anthropic: ByteDance is pre-training a roughly 10-trillion-parameter AI mannequin — greater than triple the dimensions of Moonshot’s Kimi K3 — aiming to strategy the size of Anthropic’s Mythos system, per business estimates. The challenge displays intensifying competitors amongst Chinese language AI labs to match or exceed US frontier fashions, amid earlier US accusations that Moonshot’s Kimi K3 was constructed by distilling Anthropic’s know-how.
HRIC on Twitter/X: Forbes reported on August 5 that U.S. AI knowledge labeling firms are supplying OpenAI, Anthropic, and even the U.S. navy on one hand, whereas promoting the identical units of coaching knowledge to China’s high AI labs on the opposite. For years, Washington has imposed layer upon layer of restrictions on chips, but has accomplished nearly nothing to control coaching knowledge.
Taiwan detains Nvidia employee in chip smuggling probe: supply acquainted with case: Taiwanese prosecutors detained an Nvidia worker as a part of a widening probe into smuggling roughly 50 AI servers with restricted Nvidia chips to China by way of solid paperwork, bringing the whole detained to seven folks together with Tremendous Micro and Albatron employees. Nvidia denies involvement, saying it solely sells via vetted companions and “smuggling is a nonstarter.”
China’s tech advances are inflicting chaos from Silicon Valley to the White Home: A wave of Chinese language breakthroughs—free, highly effective open-weight AI fashions like Kimi K3, mass-produced specialty chips, and superior robotics—has triggered market selloffs and break up Silicon Valley, with corporations like Microsoft and Nvidia opposing restrictions on open fashions whereas OpenAI and Anthropic name them safety dangers.
Might AI take your job? Some employees in China already know the reply: Chinese language employees, from Wuhan taxi drivers displaced by autonomous automobiles to a Beijing cinematographer changed by AI software program, describe deep anxiousness over AI-driven job losses, with some profitable court-ordered severance after AI changed them. Even China’s state media now acknowledges the “collision” between AI and conventional labor rights, signaling a shift in Beijing’s tone on the difficulty.
