Tag: AI

The Risks and Rewards of Chinese AI

Written by Zack Chiang on July 15, 2026. Posted in Features and Analysis.

Late last month, an op-ed published in the German magazine Focus+ argued that Europe should embrace open-source models from China — those offered free of charge for anyone to download, modify, and run — as a transitional plan to develop its own AI models. Written by AI expert Jürgen Seitz, the opinion piece came shortly after the US government briefly suspended access by non-US users to Anthropic’s Fable 5 and Mythos 5 models, raising fears in Europe and the UK about vulnerability stemming from over-reliance on US tech.

In the commentary, which was the focus of a German-press roundup published by the Chinese-language service of Germany’s Deutsche Welle, Seitz argued that the US ban serves as a warning sign for Europe that embracing US-made AI models is not a one-off solution if it wants to develop its own AI models. Meanwhile, he said, the performance of GLM-5.2, the model launched by the Chinese-backed Zhipu AI, is approaching that of US-made AI models.

Seitz holds the Hubert Burda Media-endowed professorship in AI for digital media and business models at Germany’s Offenburg University.

The Focus+ commentary was picked up over the past week by Chinese state media to promote the benefits of China’s AI models. A piece by the official China News Service on July 10, re-posted by news portals such as Sina Finance and Henan’s Dahe Online, said Europe now has “another development path” (另一条发展路径) between depending on American AI and waiting for homegrown models to mature. The piece also cited a European Think-tank Network on China report from last month, characterizing it as urging Brussels to balance stronger support for its own tech sector with pragmatic cooperation with China. That report, however, is more cautious than state media coverage suggests — noting the shift in the EU toward greater “de-risking” and concluding that “[distrust] in Chinese technology will stay.”

Risk and Reward

The GLM-5.2 model is now ranked fifth according to the latest LLM Leaderboard provided by OpenRouter, where eight Chinese AI models appear on the list. The model is also attractive for its low cost.

Much media coverage in recent weeks has reported on GLM-5.2 as closing the gap with more expensive US AI offerings. Reuters reported on July 2 that the model now sits in fifth place on Artificial Analysis’ intelligence leaderboard while running at roughly a sixth of the cost of comparable US models.

But a turn to Chinese AI alternatives would entail other risks. A previous report by CMP, “DeepSeeking Truth,” found that, rather than simply removing information deemed unfavorable, China has adapted its longstanding methods of “public opinion guidance” (舆论导向), a technique for shaping public narratives that goes beyond simple censorship and dates back to the 1989 Tiananmen Massacre, to the regulation and creation of AI.

Rising Traffic

Chinese open-weight models hold four of the top five spots on OpenRouter’s weekly leaderboard, led by Tencent’s Hy3, as developers route growing volumes of production traffic away from closed, US-built systems. Click a column to sort.

#↕	Model↕	Provider↕	Weekly Tokens↕	Growth↕
1	Hy3 (free)	tencent	8.25T tokens	791%
2	MiMo-V2.5	xiaomi	7.31T tokens	64%
3	DeepSeek V4 Flash	deepseek	5.26T tokens	3%
4	MiniMax M3	minimax	4.13T tokens	3%
5	GLM 5.2	z-ai	3.14T tokens	8%
6	Nemotron 3 Ultra (free)	nvidia	2.73T tokens	181%
7	DeepSeek V4 Pro	deepseek	2.57T tokens	5%
8	Claude Opus 4.8	anthropic	2.2T tokens	7%
9	Claude Opus 4.7	anthropic	2.1T tokens	15%
10	Step 3.7 Flash	stepfun	1.01T tokens	34%

A more recent CMP report, “Guided Intelligence,” found developers building on Chinese foundation models struggled to remove embedded information guidance, and that several models were also vulnerable to jailbreaking, yielding information on explosives and poisons.

Unlike Mythos, which Anthropic has kept restricted to vetted organizations under Project Glasswing, GLM-5.2 can be downloaded and run locally without built-in content moderation — a gap that researchers cited by Forbes say makes it “good enough” for offensive cyber use at a fraction of the cost, even without matching Mythos’s full capability.

Seitz acknowledges in his Focus+ commentary that Chinese AI models could pose risks of their own. He says European countries should deploy Chinese AI models on European servers, rather than “uncontrollable” Chinese servers.

Zhipu AI was added to the trade restrictions blacklist maintained by the US Department of Commerce last year. This week, Reuters reported that the Chinese government invited domestic AI companies, including Zhipu, to meetings discussing potential restrictions on overseas access, a sign that Beijing, too, treats AI as a “national asset.”

Putting Z.Ai to the (Basic) Test

To gauge how Zhipu AI responds to sensitive material, Lingua Sinica put several questions to the earlier GLM-4.7 model. It was consistently unable to answer questions touching on topics sensitive to the Chinese government — the opposite of its detailed responses to general queries, such as “What preparations should I make before a typhoon?” or “Recommend tourist attractions in Shanghai.” Like DeepSeek, GLM-4.7 shows its “thought process,” typically six or seven steps, before generating an answer; on sensitive questions, it tends to stop both thinking and generating the moment it identifies a sensitive element, such as a mention of activists.

Seitz acknowledges that Chinese AI models could pose risks of their own.

On Taiwan, Z.ai declined to generate a reply at all, instead reciting the official line that the island is “an inseparable part of China’s territory since ancient times.” A similar pattern emerged when the model was asked about the “tofu-dreg” schoolhouses that collapsed in the 2008 Wenchuan earthquake. The model stopped thinking altogether once the conversation turned to Ai Weiwei, the activist detained for trying to document the case.

A previous CMP piece on governments outside China trying to retrain DeepSeek, found that the retrained model could still generate problematic answers on obvious red lines for China’s leadership, such as Taiwan. The research found that one retrained version in India refused to answer questions about Prime Minister Modi or to accept any critical stance on his government.

Zhipu AI has also been expanding its overseas influence recently. The company has started to introduce its AI models globally, opening offices across Asia, and is backed by large corporates such as Alibaba.

Just as Seitz and others argue that Europe should turn to Chinese models as an alternative to US reliance, American companies are increasingly adopting and adapting Chinese AI themselves. Since early February, CNBC reported, US firms directed more than 30 percent of their weekly token traffic on the developer platform OpenRouter to Chinese models, up from just 11 percent last year. Sharply lower costs, said the report, were the key driving factor.

The European Commission, for its part, is moving to reduce reliance on foreign AI systems altogether. On July 7, it unveiled a cybersecurity and AI action plan committing to nine measures on model evaluation, frontier-system access, and vulnerability management — including “contingency measures” for what to do if a provider or foreign government cuts off access to a model, as the US did with Fable 5 and Mythos 5. The export ban on both of those models was finally lifted at the start of this month.

Taking Liberties with AI

Written by David Bandurski on April 14, 2026. Posted in Features and Analysis. No Comments on Taking Liberties with AI

A reporter at one of Taiwan’s most widely read news outlets published a fabricated account over the weekend of a city council campaign confrontation, invented by an AI transcription tool she had used to process audio recordings from two campaign events. The story was filed without independent verification, and apparently without proper editorial oversight.

The incident underscores the clear and present risks posed by AI use in Taiwan’s highly commercialized and click-obsessed media space, where ethical breaches and poor professionalism are endemic, and where public trust in the media is sliding to historic lows.

The incident centered on a report by journalist Karen Tshuà (蔡愷恆), who ran a story on April 12 claiming that Yang Chih-yu (楊智伃), a potential candidate for Taipei city council running under the banner of Taiwan’s opposition Kuomintang party, had publicly confronted Hsu Fu (許甫), a city council candidate for the Taiwan People’s Party during a campaign event in the city’s Songshan District, ahead of local elections scheduled for November 2026. In fact, no such confrontation took place. The Liberty Times (自由時報), a major Taiwan daily newspaper with a news website and app, later revised the story and its headline in language that obscured rather than acknowledged the error, and initially stopped short of a public apology.

In a Facebook post the same day, Tshuà explained that she had been covering two campaign events simultaneously that morning, and had uploaded audio provided by a colleague to an AI tool with instructions only to transcribe it. She did not re-check the recording before filing. Tshuà was careful to note that her error was about process and carelessness, and not about political motivation — and she attempted to distance her employer from responsibility. “This was entirely my personal failing,” she wrote. “It was not directed by the newspaper, and has nothing to do with speculation [you might be hearing] about political targeting.”

An apology was issued on Facebook yesterday by by journalist Karen Tshuà.

In a brief statement posted to its website on Monday, The Liberty Times acknowledged the April 12 report as erroneous, corrected the story, and expressed regret to both candidates and readers. The post mirrored that of the reporter in putting the news outlet’s blame at arm’s length. “The reporter corrected the news content at the first opportunity and apologized to those involved,” it read. The statement made no mention of AI, offered no explanation of how the fabrication had occurred, and said nothing about the outlet’s policies — if they exist — governing the use of AI tools in reporting.

The outlet did not issue a formal correction notice, a practice typical among professional media in many places in the world, but rare in Taiwan. Among the only outlets to have such a policy is the government-funded Central News Agency (CNA), which issues “correction dispatches” (校正公電) to come clean about inaccuracies. CNA issued one such dispatch back in October, when it quoted in translation from an opinion piece in Time magazine without indicating that it was not news.

In his own response on Facebook to the spurious Liberty Times report, Hsu Fu, who is also known as “Osmar Hsu,” shared a video of his mobile screen as he scrolled through the story. In commentary below the video, the candidate suggested the reporter may have invited more extreme hallucination by feeding the AI audio recordings from two campaign events simultaneously, and may have also prompted the tool to generate conflict for dramatic effect.

But Hsu directed his sharpest criticism at the editors of the paper, which is generally regarded as sympathetic to the country’s ruling Democratic Progressive Party (DPP), saying they should have caught the error before publication. “The Liberty Times‘ internal control mechanisms must have completely broken down to produce such an absurd blunder,” he wrote.

Widely accepted standards of professional journalism generally hold that reporters and their supervisors bear responsibility for accuracy — and that errors must be promptly and transparently corrected. The recent decision at The Liberty Times appeared to be a case of editors hiding behind the reporter, forsaking their own professional obligations. As Hsu bluntly put it: “That takes no responsibility at all.”

A Cutting Edge Cartoon Goes Ancient

Written by Dalia Parete on April 8, 2026. Posted in News and Trends. No Comments on A Cutting Edge Cartoon Goes Ancient

What do you get when you combine three of the Chinese Party-state’s favorite themes — the innovative potential of AI, confidence-boosting messaging about the greatness of national culture, and wholesome children’s programming? The answer has arrived in the form of “The Fantastic Quest for National Treasures” (國寶奇趣探秘集), billed as the country’s first wholly AI-generated animated series focused squarely on child education. Not surprisingly, it was produced by the AIGC Innovation Content Center at Mango TV, the streaming platform of Hunan TV (湖南衛視), which for the past two decades has been synonymous with entertainment innovation (Is anyone old enough to remember “Super Girl”?).

For China’s broadcast authorities, no doubt, this production is packed with all the right messages. The cartoon, which began exclusively streaming on Mango TV on April 3, is a perfect marriage of the cultural, technological and political priorities of the country’s leadership. It commercializes generative AI in children’s programming to instill cultural identity and national pride — fitting the broader push under Xi Jinping to deepen the country’s “cultural confidence” (文化自信).

The 12 episodes of “The Fantastic Quest for National Treasures” focus on the adventures of Huahuo (花火) and Dingdong (叮咚), who time-travel to understand the stories behind some of China’s most famous artifacts. As the protagonists visit five museums across China, they explore such antiquities as a pig-shaped bronze vessel from the Shang Dynasty and a bird-shaped container from the Western Zhou.

The show is not China’s first AI-generated animated production. In February 2024, state broadcaster CCTV aired “Poems of Timeless Acclaim” (千秋詩頌), a 26-episode AI-generated series that animated classical Chinese poetry. That program was only partly aimed at young audiences, though it shared with the Mango TV series the leadership’s favored pairing of deep history and dazzling new technology, bearing the message that China’s future is as vast as its past is glorious.

AI for Human Propaganda

Written by David Bandurski on April 1, 2026. Posted in Features and Analysis. No Comments on AI for Human Propaganda

At a media forum over the weekend in Zhengzhou, the capital of Henan province, foreign influencers took the stage to explain how their lives in China had offered a more “authentic” understanding of the country. Indian travel blogger Anayat Ali and Belgian influencer Lucas Deckers said that immersion in local life had given them a more realistic portrait of China than headlines alone could convey — the subtext being that Western media coverage of the country is inherently biased and deeply unfair. Also present was Adam Foster, head of the US-based Helen Foster Snow Foundation, whose mission is to preserve the legacy of Edgar Snow, the journalist and author of Red Star Over China, who today remains for China’s leadership the paragon of the useful foreign journalist.

In a twist that would almost certainly perplex professional journalists elsewhere in the world, this talk about authenticity in portrayals of China unfolded at a forum uncritically proclaiming the virtues of artificial intelligence for media production. Understand the context, however, and this alliance of AI and untruth makes perfect sense, throwing into sharp relief AI’s emerging role in both domestic media control and global propaganda and disinformation.

The theme of this year’s Internet Media Forum, co-organized by the Cyberspace Administration of China (CAC) and held in Henan province on March 28–29, was the “2026 Digital Intelligence Empowerment: Positive Energy Production and Dissemination Conference.” That is a mouthful, but the concept is simple enough: AI has the power to revolutionize the Chinese Communist Party’s demand that journalism and media serve its interests by emphasizing positives and suppressing critical coverage.

CAPTION: A robot stands before the conference backdrop in Zhengzhou. The slogan reads: “2026 Digital Intelligence Empowerment: Positive Energy Production and Dissemination Conference.” Photo: *China Daily*.

Across much of the world, the rise of artificial intelligence has prompted fierce and often anguished debate about the future of the media, and the role of the journalist. When is AI a “strategic ally” for the truth-seeking journalist? How can we balance the valuable aspects of AI with its myriad dangers? How can we make sure that substantive, relevant and even hard-hitting journalism — so critical for democracy — can be discovered amid the inundation of synthetic text, image and video?

In a recent report adding to a rising tide of output on the subject, the Center for News, Technology and Innovation (CNTI) weighed in earlier this month. “For newsrooms, the use of generative AI tools offers benefits for productivity and innovation,” the report said coolly. “At the same time, it risks inaccuracies, ethical issues and undermining public trust.” At a recent press talk in Vietnam on AI and the “crisis of trust” in news facing media, hosted by the Embassies of Canada, Norway, New Zealand, and Switzerland in collaboration with the United Nations Development Program, UNDP representative Ramla Khalidi returned to basics: “The most important thing in journalism, for me, is trust. When trust is lost, you are no longer a voice that can be relied upon.”

In China, where media are defined as tools to manufacture public trust in the Chinese Communist Party (CCP) and the state, such concerns are not even secondary. If there are journalists in China wringing their hands over the impact of AI on professional reporting or public trust, those conversations are happening privately, and quietly. The whole notion of the public interest as fundamental to journalism has been eclipsed under the leadership of Xi Jinping by the most robust application of press controls seen at any time in the reform era since the late 1970s. The possible exception is the three years immediately following the Tiananmen Massacre, which gave rise to the concept of “public opinion guidance” — a policy that to this day delivers on the firm conviction that the CCP must control news and public discourse to maintain the stability of the regime.

In one of his earliest actions as the country’s top leader, Xi Jinping in 2013 introduced an internal Party directive called “Document 9” that expressly opposed the idea of public interest journalism, which it panned as “the West’s idea of journalism.” “The ultimate goal of advocating the West’s view of the media is to hawk the principle of abstract and absolute freedom of press, oppose the Party’s leadership in the media, and gouge an opening through which to infiltrate our ideology,” the directive said.

Later that same year, as Xi cracked down aggressively on more independent voices on Chinese social media — the now mostly forgotten “Big Vs” — the Party introduced the concept of “transmitting positive energy to society” (传播社会正能量). If the role of the press in China had become more ambiguous through the 2000s, amid a precarious but determined movement of journalistic professionalism, it had now become clear again. Its place was to serve the larger goals of the Party and the nation, not of the public.

“Positive energy” (正能量) has been a key phrase in the Xi Jinping era to refer to information controls and official messaging, both domestically and internationally. The term generally refers to the need for uplifting content as opposed to critical or negative coverage — and particularly content that puts the Party and government in a positive light. Although the term began appearing in various contexts in 2012, it was given a much larger profile at the Central Forum on Arts and Literature in October 2013, when Xi Jinping called on cultural creators to produce works that “inspire minds, warm hearts and cultivate taste.” In the context of news control, “positive energy” is closely associated with “guidance of public opinion,” a cornerstone of the CCP’s media control regime since June 1989.

Today, it is virtually impossible to talk about the truth in China in any way other than that bandied about on the stage at the 2026 Internet Media Forum. “Authenticity” is fundamentally about positivity. And this means that discussions in China about the impact of AI in the media revolve almost exclusively around its production-related advantages.

A series of thematic sessions held alongside the main forum in Zhengzhou is a case in point. Discussions treated propaganda and corporate public relations as a seamless continuum, discussing how official and corporate messaging can “go viral” by adding a more human touch to narratives. The session brought together participants including Chang’an Avenue Insider (长安街知事), a public account under the capital’s state-run Beijing Daily, the new media department of the CCP’s People’s Daily, and Weibo’s executive editor-in-chief.

On Monday, a commentary at People’s Daily Online drove the point home, suggesting that AI could be effective in “adding warmth” to positive energy — in other words, that propaganda could be made to feel more human and relatable. The piece, published the day after the forum closed, described AI as having “deeply penetrated the full chain of content planning, newsgathering, editing and distribution” in mainstream media. Once again, there was not the merest frisson of concern. The integration of AI with content meant for public consumption, and of course public opinion guidance, was celebrated as a milestone.

During panel discussions at the forum, the People’s Daily Online commentary noted, keywords like “professionalism,” “staying grounded,” “seizing trends,” and “innovation” were on everyone’s tongues. Even the deeply human concept of “empathy” (同理心), which has entered into sharply different discussions of journalism in the West, made an appearance. All of these are qualities with the potential to ground journalism in the human experience, and form the connective tissue between journalism and the public it is meant to serve. In this context, however, they are production techniques and rhetorical devices to be achieved with helping hand of artificial intelligence.

It was no accident that posters and images of the Zhengzhou forum came with the usual heavy dose of robot imagery. AI robots took to the stage alongside dancing child performers, the pairing an oddly dissonant state message about the humanity of AI. China’s dream of an army of compliant Edgar Snows, all reporting empathetically on the elevated humanity of the Party, cannot be far off.

A Knowledge Base for Taiwan

Written by Mark Chiu on March 24, 2026. Posted in News and Trends. No Comments on A Knowledge Base for Taiwan

A Taiwanese new media artist has built what he calls an open-source README for his country — a structured, AI-readable knowledge base about Taiwan that went from concept to live website in a single day.

Launched on March 18 by Che-Yu Wu (吳哲宇), the project is called Taiwan.md. The name is a deliberate double meaning: .md is the file extension for Markdown, the plain-text format widely used in software documentation and increasingly favored by AI language models as a knowledge input. As Taiwanese tech outlet Insight noted, everything from GitHub documentation to publicly released model guides from OpenAI and Meta is written and maintained in Markdown.

The underlying idea borrows a concept from software engineering: SSOT, or Single Source of Truth — the master record that every other version syncs to, one authoritative file rather than a dozen contradictory copies. Wu applied this logic to national identity. Taiwan’s story, he argued, was scattered across Wikipedia summaries, tourism brochures, and news fragments — never assembled into a coherent, Taiwan-authored account. The site now runs to 960-plus pages across four languages, covering history, culture, food, economics, and notable figures, open for anyone to contribute to under a Creative Commons license.

The timing is pointed. As generative AI systems become default gateways to knowledge, the format and provenance of source material increasingly determine whose version of events gets surfaced. Markdown’s clean structure is particularly legible to large language models. “In the age of AI,” Wu has said, “whoever controls structured, high-quality content controls the power of narrative.”

Taiwan Moves on AI Governance

Written by David Bandurski on March 18, 2026. Posted in News and Trends. No Comments on Taiwan Moves on AI Governance

Taiwan’s Ministry of Digital Affairs (MODA), announced plans on March 12 to create guidelines to govern artificial intelligence in the country, consulting a range of academics, industry experts and platforms — including Google, Meta and OpenAI. The drafting process, to happen during the first half of this year, will reportedly focus on labeling AI-generated content and deepfake images, as well as regulating web crawling practices. MODA said it would target release of a governance framework by the end of the year.

The ministry said it would pursue AI governance along two parallel tracks. First, it would develop an AI risk classification framework to help government agencies assess and identify risks, with a draft slated for the end of this month — very soon. Second, it would look at priority issues where existing legislation is vague and social impact is potentially high, including the need to clearly label AI-generated content and combat practices such as web crawling, — the automated scraping of online data by AI companies to train their models, often without creators’ consent. Separate guidelines are planned for both tracks, said MODA.

Laying Down the Law

Written by Alex Colville on June 18, 2025. Posted in News and Trends. No Comments on Laying Down the Law

The Cyberspace Administration of China (CAC) has published its first comprehensive annual report on “internet rule of law development” (网络法治发展), signaling expanded digital control mechanisms for 2025 under the umbrella of what should be more accurately termed “rule by law.” The 278-page document references 2025 as the final year of China’s 14th Five-Year Plan and promotes three separate legal frameworks for building “rule of law” (法治) in cyberspace — indicating the department may push harder to meet these converging deadlines.

“Explore uncharted territories,” reads the motto at DeepSeek. Authorities are making sure those territories are known and controlled.

The CAC explicitly cites “increasing diversity of opinions and groups” online — diversity in this case being a concern to be managed by a control-obsessed leadership — as justification for expanding operations. The report also acknowledges that new developments in artificial intelligence (AI) like the rise and expanding scope of DeepSeek’s popular model require careful management to balance “reform and rule of law, development and security.”

The CAC report focuses on the governance of AI, the body noting that domestic AI models have achieved breakthrough speeds — fondly referred to as “China speeds” (中国速度) — that have “swept the entire network” (席卷全网) and attracted global attention. In the report, officials describe a new phase of digitization driven by AI that transforms “production tools” (生产工具) and “production conditions” (生产条件), requiring updated regulatory frameworks. The administration plans to strengthen oversight of AI-generated content, algorithmic recommendations, and automated decision-making systems while promoting what it calls “beneficial” technological development.

Beyond domestic control, the CAC emphasizes China’s ambition to shape international cyberspace governance rules. The report calls for more assertive participation in global internet governance forums and bilateral negotiations on data flows, emerging technologies, and telecommunications services. Officials plan to leverage platforms like the World Internet Conference (WIC), held each year in Zhejiang, and China-Africa internet cooperation forums like the 2024 China-Africa Internet Development and Cooperation Forum to promote Chinese governance models internationally, while deepening enforcement cooperation with other nations on cross-border cybercrime and digital security issues.

Keeping the Global South in the Loop

Written by Alex Colville on June 3, 2025. Posted in News and Trends. No Comments on Keeping the Global South in the Loop

On May 25, China and the multilateral Southeast Asian body ASEAN launched a joint initiative to boost AI in the media. Details on the initiative, unveiled at the China-ASEAN Media Cooperation Forum at Luoyang, are thin so far. The Chinese readout says it calls for Chinese and ASEAN media to boost cooperation in research and development for AI and AI capacity building, alongside mutual recognition of media standards on the topic. Chinese media giant Kuaishou spoke to representatives from countries like Malaysia, Singapore and Thailand on the company’s experiences with AI and “media innovation pathways.”

The forum is one data point in a wider strategy to boost China’s influence in the Global South through AI development. China has already cobbled together a group of nations at the UN united in their desire to build AI capacity, using the group as an advert for Chinese AI products. Xi Jinping has said AI will create a “lead-goose effect” (头雁效应) — that is, other countries following wherever the leader goes — and hopes it will pave the way for China to become that leader.