There is a Chinese Communist Party (CCP) plot to put millions at the mercy of the CCP using just released AI technology. This isn’t being overly dramatic.
The speed at which AI is improving is exponential as AI improves itself, and we are unprepared for this because we have never experienced anything like this before. For example, a few months ago music videos made on home computers began appearing with AI-generated people and scenes in them that were pretty impressive, but the people would sprout extra arms and fingers, food would inexplicably fly off plates into mouths and text on signage was random characters, but it was a giant leap ahead of what had come before.
There are several popular platforms for developing videos, and one of the most popular is Shanghai-based Hailuo AI. These platforms now are producing videos of people with fine details, emotive eyes and are close to being indistinguishable from real people, an incredible leap from just December.
Photo: AFP
CCP IS ABOUT POWER
The CCP’s purpose is to increase their power. It is by default in an “unrestricted war” with their own people, Taiwan and the West.
The CCP won the Chinese Civil War in large part through guile. Their propaganda, disinformation, rumors and ability to subvert the Chinese Nationalists (KMT) from the inside were all highly effective.
Photo: EPA-EFE
For their own people they have imposed a dystopia, including the “great firewall” to control information from the outside. It also exerts strict control over domestic Internet information, uses a vast surveillance camera network with facial recognition and monitors financial transactions done online. If the CCP can think of any way to impose more control over their subjects they will do it.
Abroad they traditionally used RICE (Reward, Ideology, Coercion and Ego) techniques to not only recruit spies, they have used it to win over politicians, scientists and other useful people. They have weaponized overseas Chinese community groups, taken over their media and even set up police stations around the world.
Through software like ByteDance’s TikTok they are capable of sweeping data collection, while Chinese hackers steal all sorts of information and attack online systems. Huawei used their telecommunications equipment to collect yet more.
They have worked to subvert algorithms even in foreign Web sites by flooding the Internet with disinformation and misinformation. Their infamous “little pink” and “50-cent” armies roam the Internet spreading their agenda.
MAKING PROPAGANDA
AI is taking this to an exponentially higher level.
The CCP is investing heavily in AI because it opens opportunities for the CCP’s United Front Work Department (UFWD) and Ministry of State Security (MSS) to vastly increase its power worldwide.
While Hailuo can be very useful in creating propaganda, TikTok owner ByteDance’s just released OmniHuman-1, which is explicitly for deepfakes and is shockingly good. It is able to produce videos from pictures, video and audio fed by the user to create videos realistic enough to require paying attention.
They produced some samples using American celebrities that I recognized enough to notice they weren’t quite right, but anyone less familiar would likely have been fooled. It will not be long before OmniHuman improves to the point that they are indistinguishable.
The gullible will fall for outrageous deepfakes in partisan social media, but these are pretty easy to discredit. It is the more subtle videos that are concerning because they can be used subtly to change the narrative, such as editing a video of the US Secretary of State and swap out “one China policy” for “one China principle.”
AI SECRET AGENT
The release last month of DeepSeek-R1 AI by Hangzhou-based High Flyer rightly attracted a vast amount of attention. Users amused themselves trying to get around the Chinese Communist Party’s (CCP) imposed censorship, but more alarmingly hackers discovered unprotected data ports, that data was being shared with TikTok and many reminded us that by law they must share any data with the CCP.
Perhaps intentionally to avoid widespread press scrutiny, the most powerful AI agent ever, UI-TARS, was released during the DeepSeek hoopla. AI agents by American companies require a paid subscription but offer powerful research capabilities and other functions by taking over a browser and doing work for you.
Unlike previous AI agents, UI-TARS comes in two varieties, one taking over the browser like the others, but with a second option to take over the entire computer or phone.
It can install software, scrape any bit of data it likes and make all sorts of modifications all on its own following whatever instructions it is given whether online or not. That could completely change how we work, play and communicate on our devices.
UI-TARS is open source, so unlike the American AI agents, developers can access, modify and distribute the software for free. This should encourage widespread adoption, including under different branding as long as they retain the original copyright notice, license text and notices in the source code, which non-coders never read.
Why would they do this for free instead of requiring a subscription? To make sure it gets on to as many devices as possible.
How nice of ByteDance, the developer of UI-TARS.
WORLD’S MOST DANGEROUS TROJAN HORSE
Soon people will be downloading off-brand UI-TARS without knowing it, and there could be hundreds or even thousands of brands running it. Your ACME brand AI agent running on UI-TARS can act as spyware tracking your every move and stealing all your data for Beijing, and it will know everything about you — opening up blackmail opportunities on a massive scale.
It can also act as malware to destroy your life if ByteDance is ordered to. It can destroy relationships, cause people to get fired or drain your bank account by pretending to be you.
It can even operate like the “great firewall.” By replacing your browsers with look-alikes, modify the operating system to make it look and feel like the real browser and even swap out real web address URLs to display fake ones.
This would allow ByteDance to direct you to a fake Google search engine to provide you with modified results like TikTok does, and direct you to modified replica sites. In the address bar it might read “taipeitimes.com” or “nytimes.com” but in reality could be modified versions with subtle changes parroting the CCP line.
If it is not too slow, another possibility is they might opt to keep the original browsers and sites but edit the content the same way as it is downloading. This would have the same effect, but might become obvious on slower devices.
As is the case on TikTok, results playing up the CCP line would also be prioritized and content scrubbed from the results as DeepSeek-R1 AI does now, albeit still rather clumsily. DeepSeek-generated articles and books, propaganda videos made with Hailuo AI and deepfake videos made or modified by OmniHuman would feature prominently.
Millions of people around the world could soon be constantly surveiled through their own cameras and microphones, monitored and tracked and living in an alternate information reality — just like in China.
The CCP would have the ability to control nearly every aspect of these people’s lives — just like in China.
But unlike the Chinese, they would not even know how much power they have lost to the CCP.
Donovan’s Deep Dives is a regular column by Courtney Donovan Smith (石東文) who writes in-depth analysis on everything about Taiwan’s political scene and geopolitics. Donovan is also the central Taiwan correspondent at ICRT FM100 Radio News, co-publisher of Compass Magazine, co-founder Taiwan Report (report.tw) and former chair of the Taichung American Chamber of Commerce. Follow him on X: @donovan_smith.
June 2 to June 8 Taiwan’s woodcutters believe that if they see even one speck of red in their cooked rice, no matter how small, an accident is going to happen. Peng Chin-tian (彭錦田) swears that this has proven to be true at every stop during his decades-long career in the logging industry. Along with mining, timber harvesting was once considered the most dangerous profession in Taiwan. Not only were mishaps common during all stages of processing, it was difficult to transport the injured to get medical treatment. Many died during the arduous journey. Peng recounts some of his accidents in
“Why does Taiwan identity decline?”a group of researchers lead by University of Nevada political scientist Austin Wang (王宏恩) asked in a recent paper. After all, it is not difficult to explain the rise in Taiwanese identity after the early 1990s. But no model predicted its decline during the 2016-2018 period, they say. After testing various alternative explanations, Wang et al argue that the fall-off in Taiwanese identity during that period is related to voter hedging based on the performance of the Democratic Progressive Party (DPP). Since the DPP is perceived as the guardian of Taiwan identity, when it performs well,
A short walk beneath the dense Amazon canopy, the forest abruptly opens up. Fallen logs are rotting, the trees grow sparser and the temperature rises in places sunlight hits the ground. This is what 24 years of severe drought looks like in the world’s largest rainforest. But this patch of degraded forest, about the size of a soccer field, is a scientific experiment. Launched in 2000 by Brazilian and British scientists, Esecaflor — short for “Forest Drought Study Project” in Portuguese — set out to simulate a future in which the changing climate could deplete the Amazon of rainfall. It is
The Taiwan People’s Party (TPP) on May 18 held a rally in Taichung to mark the anniversary of President William Lai’s (賴清德) inauguration on May 20. The title of the rally could be loosely translated to “May 18 recall fraudulent goods” (518退貨ㄌㄨㄚˋ！). Unlike in English, where the terms are the same, “recall” (退貨) in this context refers to product recalls due to damaged, defective or fraudulent merchandise, not the political recalls (罷免) currently dominating the headlines. I attended the rally to determine if the impression was correct that the TPP under party Chairman Huang Kuo-Chang (黃國昌) had little of a
RSS