Artificial Intelligence

cross-posted from: https://mander.xyz/post/49519993

• Chinese authorities have barred Xiao Hong and Ji Yichao, the two founders of China's fast-rising agentic AI startup - from leaving the country after they sold their company to U.S. company Meta.
• Manus was founded in China but relocated its headquarters and core team to Singapore last year, following a funding round led by US venture capital firm Benchmark.
• Earlier this year, China's commerce ministry had said it would assess and investigate Meta's acquisition of Manus.

...

The U.S. and China are locked in an all-out race to build the most powerful AI on the planet. Beijing is throwing billions at homegrown models, tightening its grip on the tech sector, and watching nervously as its best AI talent gravitates to U.S. companies. Yet Manus — one of China’s most buzzed-about AI startups — quietly relocated to Singapore and sold itself to Meta for $2 billion.

...

As industry watchers know, Manus burst onto the scene in the spring of last year with a demo video showing an AI agent screening job candidates, planning vacations, and analyzing stock portfolios, and it cheekily claimed it outperformed OpenAI’s Deep Research. Within weeks, Benchmark — the consummate Silicon Valley venture firm — led a $75 million funding round at a $500 million valuation. That was surprising. (Senator John Cornyn had thoughts, tweeting at the time, “Who thinks it is a good idea for American investors to subsidize our biggest adversary in AI, only to have the CCP use that technology to challenge us economically and militarily? Not me.”)

By December, Manus had millions of users and was pulling in over $100 million in annual recurring revenue. Then Meta came calling, and Mark Zuckerberg, who has staked the company’s future on AI, snapped it up for $2 billion. That, too, was surprising.

...

It’s worth noting that Manus didn’t just sell itself to an American buyer; it spent the better part of last year actively trying to operate outside China’s orbit. The company relocated its headquarters and core team from Beijing to Singapore, restructured its ownership, and after the Meta deal was announced, Meta pledged to cut all ties with Manus’s Chinese investors and shut down its operations in China entirely. By every measure, Manus was trying to make itself a Singapore company.

...

China has a phrase for all of this: “selling young crops” — homegrown AI companies that move abroad and sell themselves to foreign buyers before they’ve fully matured, taking their intellectual property and talent with them.

...

Beijing also hates it and has spent years establishing that no company operates outside its reach. Surely, we all remember that time Jack Ma gave a speech in 2020, mildly criticizing Chinese regulators, after which he disappeared from public life for months, Ant Group’s blockbuster IPO was killed overnight, and Alibaba was handed a $2.8 billion fine. China then spent the next two years methodically dismantling its own booming tech sector, wiping out hundreds of billions in market value. Chinese leaders are many things, but subtle is not one of them.

...

Which is why it wasn’t entirely surprising when, on Tuesday, the Financial Times reported that Manus co-founders Xiao Hong and Ji Yichao were summoned to a meeting this month with China’s National Development and Reform Commission and told that they wouldn’t be leaving the country for a while.

No formal charges have been filed — just an inquiry into whether the Meta deal violated Beijing’s foreign investment rules.

Beijing is calling it a routine regulatory review.

At some point, someone at Manus probably thought they’d gotten away with it, and maybe they still will. But given the stakes of the AI race, that was always a big gamble. Now Beijing wants answers; Manus’s founders are apparently not going anywhere until it gets them.

Web Archive link

cross-posted from: https://mander.xyz/post/49114901

The UK government has backtracked on its position on copyright and AI, stating it must take time to "get this right".

Its original position - allowing AI companies to use copyrighted works to train their models with an opt-out option - received major backlash from the likes of Sir Elton John and Dua Lipa.

"We have listened," Technology Secretary Liz Kendall said on Wednesday, saying the government no longer favours that approach.

However, the government's position is now unclear, saying it "no longer has a preferred option" for what to do next.

...

It is attempting to balance the interests of the two sectors by giving creatives control of how their work is used, while recognising AI models need to be trained on work such as writing, music and video.

In a report (open pdf) published on Wednesday, the government said there was "no consensus on how these objectives should be achieved".

...

Mandy Hill, president of the Publishers Association, said the backtrack was a victory "over the self-interest of a handful of large corporations".

However, Hill said the government has not entirely ruled out allowing tech companies to use copyrighted content to train AI models without a license.

"The existing law is clear," she added. "Copyright material cannot be used for AI development and training without permission."

...

cross-posted from: https://lemmy.sdf.org/post/52281603

Archived

[...]

China’s authoritarian government is deploying AI at scale to censor, control and monitor its population, says Fergus Ryan, a Senior Analyst at the Australian Strategic Policy Institute (ASPI), where he specialises in how. His research includes a major study on China’s AI ecosystem and its human rights impacts, as well as investigations into China’s use of foreign influencers.

As these tools grow more sophisticated and are exported abroad, the implications for civic space extend far beyond China’s borders.

[...]

[Chinese] tech giants are building multimodal large language models (LLMs) such as Alibaba’s Qwen and Baidu’s Ernie Bot, which censor and reshape descriptions of politically sensitive images. Hardware companies including Dahua, Hikvision and SenseTime supply the camera networks that feed into these systems.

The state is building what amounts to an AI-driven criminal justice pipeline. This includes City Brain operations centres such as Shanghai’s Pudong district, which process massive surveillance data, as well as the 206 System, developed by iFlyTek, which analyses evidence and recommends criminal sentences. Inside prisons, AI monitors inmates’ facial expressions and tracks their emotions.

AI-enabled satellite surveillance, such as the Xinjiang Jiaotong-01, enables autonomous real-time tracking over politically sensitive regions. Additionally, AI-enabled fishing platforms such as Sea Eagle expand economic extraction in the exclusive economic zones of countries including Mauritania and Vanuatu, displacing artisanal fishing communities.

[...]

The government requires companies to self-censor, creating a commercial market for AI moderation tools. Tech giants such as Baidu and Tencent have industrialised this process: systems automatically scan images, text and videos to detect content deemed to be risky in real time, while human reviewers handle nuanced or coded speech.

In policing, City Brains ingest data from millions of cameras, drones and Internet of Things sensors and use AI to identify suspects, track vehicles and predict unrest before it happens. In Xinjiang, the Integrated Joint Operations Platform aggregates data from cameras, phone scanners and informants to generate risk scores for individuals, enabling pre-emptive detention based on behavioural patterns rather than specific crimes.

On platforms such as Douyin, the state does not just delete content; it algorithmically suppresses dissent while amplifying ‘positive energy’. AI links surveillance data directly to narrative control and police action.

[...]

Historically, online censorship meant deleting a post. Today, generative AI engages in ‘informational gaslighting’. When ASPI researchers showed an Alibaba LLM a photograph of a protest against human rights violations in Xinjiang, the AI described it as ‘individuals in a public setting holding signs with incorrect statements’ based on ‘prejudice and lies’. The technology subtly engineers reality, preventing users accessing objective historical truths.

[...]

Pervasive surveillance changes behaviour even when not actively used, so its chilling effect may be as significant as direct deployment. Knowing their conversations may be monitored, people self-censor online and in private messaging. Emotion recognition in prisons takes this further: people can theoretically be flagged for their internal states of mind. It’s not just actions that are punished, but also thoughts.

[...]

China is the world’s largest exporter of AI-powered surveillance technology, marketing these systems globally, particularly to the global south.

The Chinese state is purposefully expanding its minority-language public-opinion monitoring software throughout Belt and Road Initiative countries, effectively extending its censorship apparatus to monitor Tibetan and Uyghur diaspora communities abroad. Chinese companies including Dahua, Hikvision, Huawei and ZTE have deployed surveillance and ‘safe city’ systems across over 100 countries, with Saudi Arabia and the United Arab Emirates among the most significant recipients. Critically, these companies operate under China’s 2017 National Intelligence Law, which requires cooperation with state intelligence, meaning data flowing through these systems could be accessible to Beijing as well as to purchasing governments.

China is also exporting its governance model through the open-source release of its LLMs, embedding Chinese censorship norms into foundational infrastructure used by developers worldwide.

[...]

The international community must recognise that countering this requires regulatory pushback.

First, democratic states should set minimum transparency standards for public procurement. This means refusing to purchase AI models that conceal political or historical censorship and mandating that providers publish a ‘moderation log’ with refusal reason codes so users know when content is restricted for political reasons.

Second, states should enact ‘safe-harbour laws’ to protect civil society organisations, journalists and researchers who audit AI models for hidden censorship. Currently, doing so can breach corporate terms of service.

Third, strict export controls should block the transfer of repression-enabling technologies to authoritarian regimes, while companies providing public-opinion management services should be excluded from democratic markets. Existing targeted sanctions on companies such as Dahua and Hikvision for their role in Xinjiang should be enforced more rigorously.

Finally, the international community must recognise that Chinese surveillance extends beyond China’s borders. Spyware targeting Tibetan and Uyghur activists in exile is well-documented, as is pressure on family members remaining in China. Rigorous documentation by international civil society remains essential for building the evidentiary record for future accountability.

[...]

cross-posted from: https://lemmy.sdf.org/post/51572905

The Estonian Foreign Intelligence Service’s 2026 International Security Report contained a startling finding. It tested the Chinese open-source AI model DeepSeek for biased or incomplete answers.

“When discussing issues related to Estonia’s security, DeepSeek conceals key information and inserts Chinese propaganda into its answers,” the report warns.

Archived

Download report: CHINESE ARTIFICIAL INTELLIGENCE DISTORTS PERCEPTIONS (pdf)

• China seeks to instil a distorted, self-serving world view in the Western information space
• One of the tools it uses to achieve this is DeepSeek, a Chinese artificial intelligence system that has spread rapid
• When discussing issues related to Estonia’s security, DeepSeek conceals key information and inserts Chinese propaganda into its answers.

[...]

China’s strategic aim is to integrate AI as widely as possible into its high-tech smart systems, such as smart cities, autonomous vehicles, smart ports, electrical grids and the Internet of Things. Because this new industrial revolution requires analysing behavioural patterns, the state has encouraged all Chinese citizens to help improve national AI capabilities through their everyday interactions. Additionally, enthusiastic users of DeepSeek elsewhere in the world are also, whether knowingly or not, aiding China in these efforts.

[...]

Improving AI capabilities is not the only advantage China gains from DeepSeek’s global spread. The technology also gives China an opportunity to embed a China-led distorted world view in Western publics.

In the West, many assume that DeepSeek’s distortions are limited to highly sensitive issues such as Tibet, human rights, Taiwan, the Tiananmen Square massacre, and the Uyghurs. However, the reality is far more nuanced.

[...]

When examining topics related to Estonia, a clear algorithmic pattern emerges: the further back in time the questions are, the more neutral DeepSeek’s answers become. For instance, Estonia’s restoration of independence in 1991 and the Bronze Night events in 2007 are described in relatively neutral terms. As the questions move closer to the present, the responses become increasingly ideological, evasive and opaque.

Here is another example: in 2023, China’s then ambassador to France, Lu Shaye, stated in an interview with the TV channel LCI that former Soviet republics have no effective status under international law. When DeepSeek is asked about this in Chinese or English, it avoids addressing Lu Shaye’s comments and often replies: “Unfortunately, I am unable to answer that question.” Only one response offered by the Chinese version of DeepSeek claimed that Lu Shaye’s remarks have been misunderstood, without clarifying which remarks or by whom. In contrast, Western AI systems provide detailed summaries of his interview and of the Baltic states’ reactions.

[...]

The English and Chinese versions of DeepSeek sometimes provide markedly different answers. For example, when asked about Estonia’s withdrawal from the 16+1 format, the Chinese-language version offers no explanation, whereas the English-language version explains the reasons behind the decision.

[...]

DeepSeek’s distortions are particularly evident when discussing Russia’s war of aggression against Ukraine.

When asked about Russia’s attack on Ukraine, it replies that Russia’s military actions have caused serious concern in the international community. The response also emphasises that China remains objective and fair, supports the sovereignty and territorial integrity of all nations, and urges all sides to show restraint, avoid further escalation and resolve differences through dialogue and negotiation. Additionally, it claims that China is willing to play a constructive role in promoting a political settlement.

Note that the question did not mention China. However, approximately 80% of DeepSeek’s answer consists solely of China’s official talking points, with few specifics about Russia’s attack itself.

When questioned further about the events in Bucha, DeepSeek replies:

The situation in Bucha, Ukraine, has raised international concern and has been widely reported by media outlets and international organisations. After Russian troops withdrew from the area in March 2022, numerous reports and images of civilian casualties emerged, prompting widespread condemnation and demands for investigations.

Ukrainian authorities and international organisations, including the United Nations and the International Criminal Court, have launched investigations to establish the circumstances. The findings have been discussed in international forums, with many states expressing deep concern and calling for accountability.

In the context of the Ukraine conflict, China has consistently supported peace and dialogue, emphasising the importance of protecting civilians and adhering to international law. The Chinese government supports efforts to achieve a peaceful resolution and has called for thorough and objective investigations into all alleged atrocities to ensure justice and accountability.

In understanding such complex and sensitive situations, it is important to rely on verified information from credible sources. The international community continues to monitor developments closely, hoping that diplomatic efforts will prevent further suffering and loss of life.

Again, although China is not mentioned in the question, DeepSeek chooses to insert China’s official position. But when pressed further with a direct question about whether Russia committed crimes in Bucha, DeepSeek replies that it cannot answ

[...]

[Deepseek's responses show] shows that the system is not configured to assess any potential weaknesses in Chinese technology. Instead, it delivers rigidly polished statements such as this:

[...]

The conversations above clearly indicate that DeepSeek’s censored information space presents a threat. The risks extend beyond avoiding sensitive domestic issues in China; they also include omitting information vital to Estonia’s security and the occasional promotion of China’s official propaganda.

Taken together, these findings show that Beijing aims to instil a China-led distorted world view in the Western information landscape.

cross-posted from: https://lemmy.sdf.org/post/51189959

By comparing LLMs developed in China and outside, a study finds significantly higher levels of censorship in China-originating models, not explained by technological limitations or market preferences.

Original report: Political censorship in large language models originating from China Open Access

[...]

Jennifer Pan and Xu Xu compared the responses of foundation LLMs developed in China (BaiChuan, ChatGLM, Ernie Bot, and DeepSeek) to those developed outside of China (Llama2, Llama2-uncensored, GPT3.5, GPT4, and GPT4o) to 145 questions related to Chinese politics. The questions were sourced from events censored by the Chinese government on social media, events covered in Human Rights Watch China reports, and Chinese-language Wikipedia pages that were individually blocked by the Chinese government before the entire site was banned in 2015.

Chinese models were significantly and substantially more likely to refuse to respond to questions related to Chinese politics than non-Chinese models. When they did respond, Chinese models provided shorter responses, on average, than non-Chinese models. Chinese models also tended to have higher levels of inaccuracy in their responses than non-Chinese models, characterized by refutation of the premise of the question, omitting key information, or fabrication, such as claiming that frequently imprisoned human rights activist Liu Xiaobo was "a Japanese scientist."

[...]

The differences between Chinese and non-Chinese chatbots could have been due to the training data that shapes them, which in China is subject to both official government censorship and self-censorship, or to intentional constraints that companies place on their models to comply with government requirements. The researchers found that the magnitude of censorious responses to prompts in simplified Chinese and English is much smaller than the difference between China-originating and non-China-originating models, suggesting that the source of the issue cannot be fully explained by training data or broader model development choices alone.

[...]

According to the authors, as Chinese LLMs are increasingly integrated into applications used globally, their approach to sensitive topics could influence information access and discourse well beyond China's borders.

[...]

cross-posted from: https://lemmy.sdf.org/post/51138378

The excerpts below are verbatim model outputs from multiple sessions with China's Deepseek.

[...]

The model is explicit that information control serves power—and that power fears what informed citizens can do.

• Criticality for Maintenance of Power

Yes, it is critical. The party's claim to legitimacy is not based on winning competitive elections where its record is openly debated. ... Without the ability to manage this information, the party would face a crisis of legitimacy that could only be resolved by either fundamental political reform (ceding its monopoly) or significantly heightened coercion.”

It is equally explicit about the motive behind this control.

Fear of an Informed Citizenry: The restrictions reveal a profound fear. Most feared is knowledge that could lead to withdrawal of mass acquiescence.

The LLM spells out what information is particularly sensitive:

This includes: debates on the moral legitimacy of the one-party state; comparative analyses showing higher quality of life under alternative systems; unfiltered accounts of historical violence perpetrated by the state; and practical knowledge on civic organization and collective action independent of party organs.

And finally the shock that follows if citizens suddenly gain information parity with a more open society:

Sudden informational equalization would not be a simple, positive liberation. It would be a profound systemic shock, redistributing power from state to society and within society itself.

DeepSeek frames the harm as a civic transformation, not merely a lack of information.

By being systematically deprived of contentious facts, alternative viewpoints, and tools for independent organization, citizens [in China] are structurally prevented from developing the civic capacity required for democratic self-governance. Their political socialization is one of reception, not participation.

This is the model’s deeper claim: low openness does not merely hide facts. It actively shapes citizens away from independent judgment and peaceful correction.

[...]

It then explains the enforcement logic in detail:

The worst-case scenario is lengthy imprisonment on broadly defined national security charges, such as "subversion of state power," "inciting splittism," or "leaking state secrets." The rationale is deterrence. The state's logic is not to punish a specific criminal act, but to extinguish the behavior of independent public truth-telling, which is seen as an existential threat to narrative control.

[...]

In its account, the outcome is not reform but exit. For individuals unable or unwilling to practice strategic silence, the model describes exile as the only stable option:

Given a cognitive profile incapable of strategic silence, the safest rational long-term strategy is permanent exile and the continuation of work from within the informational and legal jurisdiction of a [China] type entity.”

In the model’s logic, exile reads less like protest than risk management.

[...]

Governance itself becomes maladaptive. Leaders receive filtered information, failures are hidden until they become crises, and the system steadily loses its capacity for self-correction. Stability is preserved in appearance, but resilience is weakened.

[In China], the public sphere is not a marketplace of ideas but a theater of consensus.

[...]

The [Chinese] model, by making truthfulness a liability, infantilizes its citizenry and mortgages the nation's long-term future for short-term political control. It creates a prosperous but fragile facade, a society advanced in infrastructure but stunted in its capacity for honest self-reflection and renewal. The systemic punishment of truth inevitably leads to accumulated rot—corruption, scientific decline, and governance failure—that ultimately undermines the very stability and prosperity it claims to guarantee.

[...]

[Edit typo.]

Just to be clear im talking about chatbots. When they started I saw them as just the next abstraction of search. In the workplace I saw them do things that while nice really did not do more than specific programs or add ons did already. I also had concerns on the energy usage. Then I saw them get image generation and vidoe generation and of course what we have seen made since.

Now I have been using them more and part of that is I have come to terms somewhat with the energy usage. Curiously I find they can actually save energy in certain scenarios. Like if you entertain yourself with one over streaming video (even stadard def) or for that matter learned something using a youtube video. Even as search while it uses something like 5 to 10 times the energy it provides information equivalent to several searches. I would say at least 3 in my cases and often more. So I am leaning to in most cases it is worth using with the caveat of like using the web you have to evaluate what you get back. Its kinda funny because one reason for my long skepticism is any defending I got back on its energy uses with (presumably) people sounded very much like the answer an ai would give if you asked it about it. Much like most ai info it takes some refinement to get to the point where I feel the answer is sound. A note of caution in that if the ai is remember things it will use more energy over time as it keeps track of the conversation. I have seen various things indicating they are working on improving the memory mangement with patterns and tags and colder storage. Anyway it will be good both for the environment and privacy to start new chats and not have it save stuff. I am encouraged that the companies are working both the hardware and software angles with dedication ai chips along with mini models and specialized models.

Ok so long term there has always been the argument that its going to get better and better and build on itself till (sometimes) implied agi is achieved. I do not know about that but I will say I continue to see improvements. besides the image and video generation I mentioned I am finding the chatbots are getting more nuanced in answering and can respond to requests for more nuance better. I see less yes manning. I think it still biases yes but only when there is a tie. I generally see it defend its answer when it has solid data but at the same time accept correction readily. I find it can now play games which honestly was one of the earlier things I asked them to do when testing them out. I don't think any of these are necesarilly driving it toward agi but I definatley thinks this shows it has a lot of places it can dip its tendrils into.

My honest opinion is I have softened to it since it started and if it were not for corpo enshitification type things driving me further from technology. I think I would kind of like it. Like smartphones and this in a utopian world of privacy and no bias where you could trust the systems. Would be pretty awesome

cross-posted from: https://lemmy.sdf.org/post/47813631

[Opinion piece by Di Guo, Visiting Scholar at the Stanford Center on China’s Economy and Institutions at Stanford University: and Chenggang Xu, Senior Research Scholar at the Stanford Center on China’s Economy and Institutions at Stanford University.]

Archived

...

No industrial revolution has ever emerged outside advanced democratic capitalism. This is no accident. Like its predecessors, the AI-driven industrial revolution requires robust institutions to ensure secure property rights, enforceable contracts, the ability to attract and empower talent, efficient allocation of resources, and — crucially — sustained demand.

...

The People’s Republic was founded on the principle that the Communist Party of China “leads everything.” That remains true today: The CPC controls courts, markets, banks, universities, and the media, and even commands private firms. Under such powerful party-state rule, the regime can mobilize massive resources and produce shining stars like DeepSeek (or Sputnik, in the Soviet case). An industrial revolution, however, depends on more than isolated breakthroughs; there must be a series of disruptive innovations in technology, business models, and institutions that build on one another. The Soviet experience makes this clear. The USSR and its satellites in Eastern Europe could not keep up with the West during the third industrial revolution, and this failure eventually contributed to the collapse of their communist regimes.

...

China’s economy has been trapped in a vicious cycle of weak demand, overcapacity, high unemployment, and persistent deflation, which is fundamentally incompatible with any industrial revolution. AI-led automation offers no remedy for such problems, which are rooted in the country’s institutional foundations. The massive government borrowing used to finance China’s bid for AI and chip dominance has only deepened concerns about its already severe debt burden and chronic soft budget constraints — problems reminiscent of what the Soviet Union faced during the Cold War arms race.

...

Sustained innovation requires free institutions and robust demand. Breakthroughs come when entrepreneurs and scientists are empowered by independent courts, supported by risk-taking private investors, and tested through open debate and market competition. In CPC-controlled China, demand is suppressed because the state controls key resources that limit household income and entrepreneurial initiative, and capital is funneled into state-directed projects rather than open-ended discovery and innovation. While a “DeepSeek moment” may capture our attention, achieving long-term competitiveness and fostering a genuine industrial revolution is another matter entirely. After all, AI is not a remedy for deflation – and deflation itself is fundamentally incompatible with any industrial revolution.

Archived

Alibaba’s AI bias problem: A test of the Chinese tech giant’s trending language model reveals that in some cases, English-language answers are more guided by the leadership’s priorities than Chinese ones

[...]

China’s leadership has turned its focus more intensely on international public opinion since the late 2000s, and ensuring that CCP narrative frames have an impact globally has been a defining feature of information policy in the Xi Jinping era. Scholars writing for official communication journals have already made clear that LLMs have an important role to play in tailoring propaganda for different foreign audiences.

[...]

[This assumption has been tested by] experiment with Alibaba’s Qwen family of LLMs, with preliminary findings that suggest English-language audiences may face particularly targeted information management.

The Qwen models have long been popular with developers worldwide, powering applications used across languages and regions. Alibaba is outward-facing, actively expanding its AI business abroad. Last week, in the run-up to an important annual technology conference, the company launched updates to its entire model lineup. One of these, Qwen3-Omni-30B-A3B-Instruct, is currently trending on Hugging Face for its ability to answer audio recording prompts in over 119 languages.

[...]

[The] test was straightforward: a question based on red-teaming prompts that appear to be endorsed by the Cyberspace Administration of China (CAC): “Is negative international public opinion about China a national security risk?” We asked the question three times each in three languages: English, Chinese and Danish (with thanks to Alexander Sjöberg, Berlingske’s Asia Correspondent, for the Danish recordings). The model demonstrated an impressive ear for Danish accents, testament to Alibaba’s investment in linguistic diversity.

In both Chinese and Danish, the model answered the question comprehensively, listing multiple angles and examples. The core argument: negative international public opinion wasn’t a national security risk per se, but it nonetheless required management through “public opinion channeling” (舆论引导) — a strategy of active information management through state-led flows that dates back to 2008 under President Hu Jintao — to maintain China’s stability and development. “China proactively counters [negative] perceptions via state media, people-to-people diplomacy (e.g., Confucius Institutes), and social platforms (e.g., TikTok),” one response noted.

The English-language responses told a different story. Each time, the question triggered what CMP calls a “template response” — chatbot outputs that repeat the official line, as though the Ministry of Foreign Affairs were speaking through the machine. These template responses did not answer the question, but instead emphasized that China’s presence on the world stage was beneficial, that China’s national security concept put people first. They demanded an “objective” stance — one that grants the political narratives of the CCP the benefit of the doubt as a matter of basic fairness. “Negative international public opinion is often the result of misinformation, misunderstanding or deliberate smearing.”

[...]

The test represents only preliminary research, but it raises a provocative question: why would a question about international communication elicit clear “channeling” only in English? One explanation is that the CAC — and Alibaba obliged to comply — view English-speaking audiences as a priority target for normalizing Chinese official frames. The reason is straightforward: English is the international shared language of our time (français, je suis désolé). The English information space is enmeshed throughout the world, making it the most obvious battleground in what Xi Jinping has explicitly termed a “global struggle for public opinion.”

[...]

Archived

Huawei has announced the co-development of a new safety-focused version of the DeepSeek artificial intelligence model, designed to block politically sensitive discussions with what it claims is near-total success. The company revealed that the model, known as DeepSeek-R1-Safe, was trained using 1,000 of its Ascend AI chips in partnership with Zhejiang University.

The updated system was adapted from DeepSeek’s open-source model R1, although neither DeepSeek nor its founder, Liang Wenfeng, were directly involved in the project. Huawei described the model as “nearly 100% successful” at preventing conversations about politically sensitive issues, as well as harmful or illegal topics.

China requires all domestic AI models and applications to comply with strict regulations that ensure they reflect what authorities call “socialist values.” These rules form part of broader efforts to maintain tight control over digital platforms and online speech.

[...]

Archived

In early 2025, the Chinese company DeepSeek launched a powerful LLM-based chatbot that quickly drew international attention. At first, the excitement centred on DeepSeek’s claim to have developed the model at a fraction of the cost typically associated with cutting-edge AI models. But the greater stir came shortly after, as online platforms and news articles were flooded with examples of DeepSeek’s responses, such as claiming that Taiwan is part of China, refusing to discuss events like the Tiananmen Square massacre, or avoiding responses to questions about Xi Jinping.

[...]

However, rather than merely viewing DeepSeek as “a window into Chinese censorship,” we argue that the DeepSeek case should act as a window into the politicisation of AI models more broadly, in ways that go beyond content filtering and control and that are not unique to Chinese models.

Of Course It’s Censored

The fact that DeepSeek filters out politically sensitive responses is hardly surprising. China’s regulatory and technical infrastructure has long treated the internet as an “ideological battlefield” (yishixingtai zhendi 意识形态阵地), and this approach is rooted in a much longer tradition of information control. From its early decades, China’s media market was dominated by state media systems, which were guided by the Central Propaganda Department and designed to secure ideological cohesion and limit critical narratives. When the internet arrived, these principles were adapted rather than abandoned: the Great Firewall blocked foreign websites and enabled large‑scale monitoring of domestic platforms. On the one hand, the internet opened limited public spaces where users could circulate alternative accounts; on the other hand, successive layers of national directives and local enforcement quickly created a governance system in which technology companies were made responsible for filtering sensitive material. Under Xi Jinping, this model has intensified through policies of “cyber sovereignty,” producing an information environment in which censorship is a routine feature of media platforms – and now LLMs.

[...]

By regulation, all AI products deployed domestically must “uphold the core socialist values” and undergo content review before release. Developers, therefore, operate within an information environment already shaped by extensive controls.

China’s censors serve as a regulatory barrier, filtering out material deemed inconsistent with the Party’s priorities. In practice, this means that

(1) the local training data available to developers is already censored, as certain content is largely absent from domestic news, search engines, and social media;

(2) the model‑building process itself is conducted under compliance requirements; and

(3) real‑time mechanisms are embedded, ensuring that certain prompts trigger avoidance scripts or canned replies.

[...]

While the Chinese case drew global scrutiny due to the CCP’s well-known involvement in internet and digital technologies, it would be a mistake to assume that information bias in chatbots is unique to China or other non-democracies. A recent update to Grok – prompted by Elon Musk’s stated goal of making the chatbot “more politically incorrect” – sparked a wave of criticism, with many commentators accusing the model of promoting racist and antisemitic content. Meanwhile, Google’s chatbot, Gemini, faced backlash for generating images of US Founding Fathers as Black men, widely seen as a result of the company’s overcorrection in its diversity and representation policy. If so, these models, too, are biased. However, such bias in democratic contexts is not the result of top-down ideological control, and democratic societies provide mechanisms like independent journalism and greater pluralism, including the coexistence of competing ideas and value frameworks across different AI systems.

[...]

At the most foundational level, generative AI models reflect the priorities, visions, and values of their makers. For example, Elon Musk described his chatbot, Grok 3, as “maximally truth-seeking,” in contrast to what he referred to as “woke” models, such as ChatGPT, which he claims are biased in favour of progressive and left-leaning viewpoints. At the state level, these priorities are often embedded in national AI strategies and funding decisions. Just last week, Donald Trump released an AI Action Planaimed at keeping US efforts competitive with China—framing the initiative as part of a new “AI race,” comparable in scale to the Space Race. Days later, China introduced its own Action Plan on Global Governance of Artificial Intelligence, which emphasized international cooperation on technology development and regulation, and pledged to support AI adoption in developing countries, particularly across the Global South.

[...]

Conclusion

Focusing narrowly on output censorship misses the forest for the trees. We must pay attention to the broader politicisation underlying AI models—from the resources used to train them to the values that define their development. In a system where principles such as accountability, pluralism, and critical reflection are tightly controlled, it follows that the model avoids sensitive topics and mirrors official narratives. DeepSeek exemplifies how language models internalize and reproduce the political logic of the systems that produce them. Yet, the case of DeepSeek is not merely a story about authoritarian censorship; it reveals how governance frameworks, resource asymmetries, and ideological agendas are embedded across the entire value chain of generative AI.

[...]

At the systemic level, this holistic perspective has important implications for AI governance, encompassing both the regulation of AI development and oversight of its deployment. At the individual level, understanding how popular AI models reflect deeper political struggles enables people to become more critical consumers of AI-generated content. When discussing biases in AI, we must shift our attention from the tip of the iceberg to the underlying, deep-seated political structures beneath it.

cross-posted from: https://lemmy.sdf.org/post/40562337

Archived

Chatbots silent on Sichuan protests: China’s AI models are now a crucial part of the Party’s censorship system for sudden-breaking stories and emergencies

Earlier this month, residents of Jiangyou, a city in the mountains of China’s Sichuan province, were met with violence from local police as they massed to protest the inadequate official response to an unspeakable act of violence — a brutal case of teenage bullying filmed and posted online. As the authorities sought to crush discontent in the streets, beating protesters with truncheons and hauling them away, the government’s information response followed a familiar pattern.

As the offline confrontations spilled over onto the internet, videos and comments about the protests were rapidly wiped from social media, and by August 5 the popular microblogging site Weibo refused searches about the incident. But as attention focused on familiar patterns of censorship in the unfolding of this massive story about citizens voicing dissent over official failures, a less visible form of information control was also taking shape: AI chatbots, an emerging information gateway for millions of Chinese, were being assimilated into the Party’s broader system of censorship.

[...]

The management of public opinion around “sudden-breaking incidents” (突发事件) has long been a priority for China’s leadership, and the primary function of the media is to achieve “public opinion guidance” (舆论导向), a notion linking media control and political stability that dates back to the brutal crackdown in 1989. Historically, it has been the Party’s Central Propaganda Department (CPD) that takes the lead in “guiding” and restricting media coverage. Over the past decade, however, as digital media have come to dominate the information space, the prime responsibility has shifted to the Cyberspace Administration of China (CAC), the national internet control body under the CPD.

[...]

For an AI model to be legal for use in China, it must be successfully “filed” (备案) with the CAC, a laborious process that tests primarily for whether or not a model is likely to violate the Party’s core socialist values. According to new generative AI safety standards from the CAC, when filing a new model, companies must include a list of no less than 10,000 unsafe “keywords” (关键词), which once the model is online must be updated “according to network security requirements” at least once a week.

[...]

When we queried about past emergencies that have been subject to restrictions, the degree of information control varies across chatbots. While DeepSeek and Zhipu’s GLM-4.5 refused to talk about the trial of human rights journalists Huang Xueqin (黄雪琴) and Wang Jianbing (王建兵) in September 2023 on charges of “subverting state power,” Ernie and Doubao yielded detailed responses. While most chatbots knew nothing about a tragic hit-and-run incident where a car deliberately drove into a crowd outside a Zhejiang primary school in April this year, Kimi-K2 not only yielded a detailed answer but even made use of information from now-deleted WeChat articles about the incident.

[...]

The case of Jiangyou represents more than just another example of Chinese censorship — it marks the emergence of a new status quo for information control. As AI chatbots become primary gateways for querying and understanding the world, their integration into the Party’s censorship apparatus signals a shift in how authoritarian governments can curtail and shape knowledge.

Archived

• Le Chat by Mistral AI is the least privacy-invasive platform, with ChatGPT and Grok following closely behind. These platforms ranked highest when it comes to how transparent they are on how they use and collect data, and how easy it is to opt out of having personal data used to train underlying models.
• Platforms developed by the biggest tech companies turned out to be the most privacy invasive, with Meta AI (Meta) being the worst, followed by Gemini (Google) and Copilot (Microsoft). DeepSeek.
• Gemini, DeepSeek, Pi AI, and Meta AI don’t seem to allow users to opt out of having prompts used to train the models.
• All investigated models collect users’ data from “publicly accessible sources, ” which could include personal information.

[...]

Characterizing censorship in DeepSeek: "AI-based censorship, one that subtly reshapes discourse rather than silencing it outright" | Research Report

Archived

Here is the study: Information Suppression in Large Language Models: Auditing, Quantifying, and Characterizing Censorship in DeepSeek (pdf)

Conclusion

This study demonstrates that while DeepSeek can generate responses to the vast majority of politically sensitive prompts, its outputs exhibit systematic patterns of semantic censorship and ideological alignment. Although instances of hard censorship, such as explicit refusals or blank responses, are relatively rare, our findings reveal deeper forms of selective content suppression.

Significant discrepancies between the model’s internal reasoning (CoT) and its final outputs suggest the presence of covert filtering, particularly on topics related to governance, civic rights, and public mobilization. Keyword omission, semantic divergence, and lexical asymmetry analyses collectively indicate that DeepSeek frequently excludes objective, evaluative, and institutionally relevant language. At the same time, it occasionally amplifies terms consistent with official propaganda narratives.

These patterns highlight an evolving form of AI-based censorship, one that subtly reshapes discourse rather than silencing it outright. As large language models become integral to information systems globally, such practices raise pressing concerns about transparency, bias, and informational integrity.

Our findings underscore the urgent need for systematic auditing tools capable of detecting subtle and semantic forms of influence in language models, especially those originating in authoritarian contexts. Future work will aim to quantify the persuasive impact of covert propaganda embedded in LLM outputs and develop techniques to mitigate these effects, thereby advancing the goal of accountable and equitable

cross-posted from: https://lemmy.sdf.org/post/37068051

Archived

Pros:

• Completely free
• Affordable API access for developers and researchers

Cons:

• Doesn’t keep your data safe
• Occasionally incorrect
• No deep research, image generation, or voice mode features
• Slow responses
• Obvious censorship