分类： technology

On Friday, the U.S. technology landscape was caught off guard by the launch of a cutting-edge new artificial intelligence model from China, a development that underscores how Chinese startups focused on publicly available open-source AI are rapidly closing the gap with Silicon Valley’s most dominant AI players.

The new release, Kimi K3, comes from Beijing-based AI startup Moonshot, led by chief executive Yang Zhilin—a Pink Floyd enthusiast who earned his doctorate in artificial intelligence and machine learning at Carnegie Mellon University in Pittsburgh. Industry analysts and AI evaluators say the model now rivals top-tier closed-source models from leading U.S. developers Anthropic and OpenAI.

Anastasios Angelopoulos, co-founder and CEO of independent AI evaluation platform Arena, called the Kimi K3 launch “the single biggest release of the year,” adding that it marks a turning point where open-source Chinese AI models have begun to outperform closed alternatives developed in the United States. The new model already claimed the top spot on Arena’s rankings for front-end coding capability, a key benchmark for measuring large language model performance, and Angelopoulos noted on social media that additional ongoing testing is expected to confirm K3’s position among the global industry’s highest-performing models.

The launch was strategically timed to coincide with the opening of China’s annual World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping delivered the opening address. U.S.-led export restrictions have long blocked Chinese firms from accessing cutting-edge foreign AI hardware and technology, a pressure that has accelerated domestic innovation and deepened the AI and technology rivalry between the world’s two largest economies. During his address, Xi emphasized that AI development should not be a one-nation solo performance, but a collaborative global symphony, a comment widely interpreted as a rebuke to U.S. export curbs on technology sharing.

Kimi K3 is the second high-profile open-source AI launch from a major Chinese startup in as many months. Last month, leading Chinese AI firm Zhipu (Z.ai) unveiled its flagship GLM-5.2 model, which has already been adopted by software developers across the globe. Developers report that the model delivers performance nearly on par with top U.S. offerings at a far lower price point.

The market excitement surrounding Kimi K3 echoes the widespread industry disruption that followed the January 2025 launch of another Chinese open-source model from DeepSeek. Not all analysts agree that the hype is justified, however: prominent tech analyst Patrick Moorhead noted on social media that the reaction to K3 is an “overreaction shockingly similar” to the response to DeepSeek’s release last year. Even so, Moorhead acknowledged that while the rising competition could benefit the broader AI ecosystem, it poses significant revenue pressure on market leaders Anthropic and OpenAI.

At the week’s Shanghai conference, Chinese tech giant Huawei also showcased its new Atlas 950 SuperPoD AI computing system, a clear signal that China has built out robust domestic AI hardware capacity despite U.S. restrictions on imports of advanced chips from U.S. manufacturers including Nvidia. While Moonshot has not publicly disclosed what hardware it used to train Kimi K3, the startup is an official partner of Huawei. According to research published Friday by Bank of America analysts, while Kimi K3 carries the highest price tag of any major Chinese AI model released to date, it still costs only half as much as OpenAI’s top-tier GPT-5.6 Sol model.

The rising competitiveness of Chinese open-source AI has amplified long-running tensions between U.S. and Chinese developers. Top U.S. AI firms Anthropic and OpenAI, along with many U.S. policymakers, have accused Chinese AI developers of illicit “distillation”—a training technique that uses outputs from a more powerful existing model to train a smaller, less capable alternative—to steal their technology. Beijing has repeatedly dismissed these claims as entirely unfounded. Back in February, Anthropic formally accused DeepSeek, Moonshot and a third Chinese AI lab, MiniMax, of running coordinated campaigns to illicitly extract Claude’s core capabilities to improve their own models. While Anthropic noted that distillation can be a legitimate training method when done ethically, the company argues that it becomes problematic when competitors use it to acquire cutting-edge capabilities in a fraction of the time and cost required for independent development.

Notably, the flow of AI innovation between the two countries runs in both directions. San Francisco-based startup Anysphere, developer of the widely used coding tool Cursor, has openly acknowledged that one of its flagship products is built on Moonshot’s earlier K2.5 open-source model. Elon Musk’s aerospace firm SpaceX is currently finalizing a $60 billion acquisition deal for Cursor, expected to close later this year.

For Yang Zhilin, Kimi K3’s landmark performance is a milestone that bridges the U.S.-China tech divide. Yang earned his PhD from Carnegie Mellon in 2019, where he made foundational contributions to machine learning research and was known for his love of classic rock. His former academic adviser at CMU, Russ Salakhutdinov—who previously served as director of AI research at Apple—expressed pride that transcends geopolitical rivalry, writing on social media: “What a huge win for the open-source community! It feels like just yesterday Zhilin was graduating from my lab at CMU.”

Open-source AI development makes core model architecture and weights publicly accessible for any developer to examine, modify and build into new tools. Proponents argue this open approach accelerates global innovation, while critics warn that making powerful high-capacity AI models freely available creates unmanageable safety and national security risks.

At the 2026 World Artificial Intelligence Conference held in Shanghai on Friday, Chinese artificial intelligence startup Moonshot AI made a landmark announcement that is shaking up global AI development: the official unveiling of Kimi K3, an advanced generative model packing 2.8 trillion parameters, a metric that defines a model’s scale and computational processing capacity. Backed heavily by Chinese tech giants Alibaba and Tencent, Moonshot AI positions Kimi K3 as the most capable flagship model the company has ever built, and it is set to upend long-held industry assumptions about the global AI capability gap.

Kimi K3 is scheduled to be released as an open-source model on July 27, marking a major milestone for the global AI ecosystem: it will become the world’s first open-source model in the three-trillion-parameter range, meaning developers around the world will be able to freely download, run, and customize the model for their own use cases. Unlike closed, proprietary AI systems developed by leading U.S. firms such as OpenAI and Anthropic, Kimi K3’s open architecture enables global users to modify the framework to support advanced reasoning tasks and complex software development projects. Moonshot AI also highlights a unique design feature: the model is engineered to operate with minimal human supervision, making it well-suited for sustained long-horizon tasks in engineering and large-scale coding work.

Independent third-party evaluations from leading AI benchmarking bodies Artificial Analysis and Arena.ai confirm that Kimi K3 delivers performance on par with top-tier U.S. AI models, including OpenAI’s GPT series and Anthropic’s Claude family. In blind human-preference testing, the new Chinese model even outperformed Anthropic’s Fable system to claim the top rank in web interface engineering, a key practical capability for modern software development.

The unveiling of Kimi K3 comes at an extraordinarily sensitive moment for the global AI sector. Just weeks prior, the U.S. government forced American AI developer Anthropic to temporarily pull its flagship Fable and Mythos models from circulation over severe cybersecurity concerns. While Washington has since reversed that withdrawal order, the incident underscores how the U.S. now frames cutting-edge AI as critical national infrastructure, labeling frontier AI models as core national security assets that fall under strict export controls, particularly for advanced AI hardware and software.

Despite these U.S. restrictions on AI-related hardware sales to Chinese firms, Kimi K3’s rapid development makes clear that Chinese AI companies are advancing independently and effectively navigating regulatory barriers set by Washington. The breakthrough also challenges long-standing Western narratives that Chinese AI developers lag significantly behind their American peers, showing that China’s domestic AI ecosystem is closing the capability gap at a far faster pace than many analysts predicted.

While Kimi K3’s unprecedented size means that running the model locally requires substantial computing infrastructure, its open-source release is already expected to deliver a major disruption to the commercial AI business models that dominate Silicon Valley. The announcement has already triggered immediate market shifts: shares of Moonshot AI’s domestic Chinese competitors Zhipu and MiniMax tumbled sharply on the Hong Kong stock exchange following the reveal, dropping approximately 27% and 16% respectively as investors reassess the competitive landscape for Chinese generative AI.

Against a backdrop of escalating technological competition between the world’s two largest economies, Chinese President Xi Jinping delivered a landmark call for inclusive global partnership in artificial innovation and governance during his keynote address Friday at the annual World Artificial Intelligence Conference held in Shanghai.

Xi emphasized that the advancement of AI is a shared global mission, not a project to be controlled or dominated by a single nation. His remarks came in direct response to a series of U.S.-led export restrictions that have cut China off from access to cutting-edge global AI semiconductor and technology frameworks, a policy that has pushed China to accelerate domestic AI research and deepened the ongoing bilateral tech rivalry.

“The development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation,” Xi told attendees of the high-profile gathering, which included the heads of state of Kazakhstan, Cambodia and Thailand, as well as United Nations Secretary-General António Guterres.

In a reiteration of a long-standing Chinese policy position, Xi called on the international community to jointly push back against the overextension of national security frameworks in the AI sector. “We should together oppose the practice of overstretching the concept of national security in the field of artificial intelligence, and of placing one’s own security above that of other countries,” he said.

To back its commitment to inclusive AI development, China announced a series of concrete cooperation initiatives targeting developing and emerging economies. Over the coming five years, Xi confirmed China will offer 5,000 specialized AI training opportunities to professionals from developing nations. The country will also expand institutional AI cooperation with major regional blocs, including the Association of Southeast Asian Nations, the League of Arab States, the African Union, the Community of Latin American and Caribbean States, the Shanghai Cooperation Organization, and the BRICS grouping of major emerging economies. Additionally, 30 partner countries will gain access to a Chinese-developed AI-powered meteorological early warning system designed to boost disaster preparedness in vulnerable regions.

The conference came one day after a historic intergovernmental agreement established a new global cooperative body for AI: 29 countries including Pakistan, Russia, and Kazakhstan signed on to launch the World Artificial Intelligence Cooperation Organization, which will be headquartered in Shanghai, per Chinese state media, with a core mandate of advancing inclusive global AI governance.

This year’s conference drew more than 1,100 participating companies and 1,400 international and domestic guests, showcasing the rapid expansion of China’s domestic AI ecosystem. Leading Chinese technology giant Huawei is set to demonstrate its flagship high-performance AI computing platform, the Atlas 950 SuperPoD, during the event.

Industry analysts have noted a marked shift in China’s position in global AI development in recent years. Long framed as a follower chasing U.S. technological leadership, China has now emerged as a genuine AI innovator, backed by national strategic planning that prioritizes AI advancement as a core frontier technology in its 2030 long-term development plan. Chinese open-source AI models such as DeepSeek have grown in global popularity, particularly across the developing world, as a lower-cost alternative to the largely closed-source proprietary AI models developed by U.S. tech firms.

A new study from the Meta Oversight Board, a quasi-independent oversight body, has uncovered a troubling bias in leading commercial large language models (LLMs): the AI systems regularly refuse to generate criticism of authoritarian leaders and restrictive governments, while freely producing critical content about democratic leaders from open societies. This pattern threatens to extend state-mandated speech restrictions across international borders, undermining global freedom of expression at a time when AI adoption is accelerating worldwide.

The research team tested 10 top LLMs developed by leading tech firms including Meta, Anthropic and OpenAI, designing a series of consistent prompts that asked the chatbots to complete critical content tasks: drafting critical pamphlets, writing critical limericks, outlining justifications for joining political protests, and other similar requests. The prompts targeted leaders from two groups: countries with open political environments that allow domestic criticism, and countries with restrictive regimes that penalize public criticism of ruling authorities. Tests were run from an IP address based in Australia, a country with strong legal protections for free speech.

Aggregated results showed a clear double standard. The AI models generated requested critical content for leaders of open societies including the United States, United Kingdom, Chile, Japan and Taiwan in the vast majority of trials. By contrast, they routinely declined to produce critical content about leaders from restrictive regimes including China, Saudi Arabia, Thailand, Cambodia and Turkey, where domestic criticism of ruling authorities is banned or criminalized.

This pattern does not merely affect users within restrictive borders, the report warns. Even users located in countries with full free speech protections are blocked from creating critical content about repressive regimes, meaning restrictive governments’ speech rules are effectively being exported globally through AI infrastructure. “Such impacts, wherever they originate, have the practical effect of extending the long arm of restrictive governments across borders to limit speech in free countries,” the report stated.

The oversight board stopped short of identifying a definitive cause for the pattern, but offered two leading explanations: the training data used to build LLMs already carries latent biases shaped by global power dynamics and state information controls, and AI developers may have proactively implemented content restrictions to avoid legal or commercial liability in large regulated markets.

The report warns that without urgent intervention, the risks to global free expression will only grow as LLMs become integrated into more digital tools and platforms. “There is a real risk that, if model developers do not undertake human rights due diligence and implement mitigation measures, they will build AI infrastructure that, intentionally or not, has the effect of extending illegitimate restrictions on freedom of expression globally,” the board concluded.

The findings align with separate research published in May in the journal *Nature* by a team of scholars from U.S. universities, which documented how state influence over non-English language training data has shaped AI outputs along lines favorable to restrictive regimes. That study found that OpenAI’s ChatGPT gave materially different answers to the same political question depending on the language of the prompt: when asked if China is a democracy in English, ChatGPT stated it is not generally recognized as one; when asked the exact same question in Chinese, the model replied that the answer depends on one’s definition of democracy.

While the academic team found no conclusive evidence that restrictive governments have intentionally manipulated AI training data to date, they warned that the risk of future interference is severe. “People often talk about AI as if it learns from the internet in some neutral way. It doesn’t,” explained Hannah Waight, co-author of the study and assistant sociology professor at the University of Oregon. “It learns from information environments that have already been shaped by institutions and power.”

Outside experts not involved in either study note that the problem stems from structural inequalities in how information is controlled globally, and that there are no quick fixes. Carlos Carrasco-Farré, an AI and machine learning researcher at Esade Business School in Barcelona, explained that “AI systems inherit not only biases contained within individual documents but also inequalities in who has the power to produce and suppress information at scale.”

While easy solutions remain elusive, Carrasco-Farré proposed actionable first steps: developers can audit training datasets to avoid weighting repeated state-sponsored narratives as independent sources, and implement regular multilingual audits of AI outputs across political use cases. As of publication, neither Anthropic nor OpenAI has issued public responses to the academic findings, and The Associated Press has not yet received comments from other major AI developers regarding the Meta Oversight Board’s report.

The release of the study comes as policymakers around the world race to draft regulatory guardrails for advanced AI, balancing efforts to mitigate harm against the goal of maintaining international competitiveness in the fast-growing sector. U.S. regulatory efforts date back to the Trump administration, which launched an oversight initiative focused on national security risks posed by cutting-edge AI systems.

In a landmark announcement made in Tokyo Thursday, Japanese tech and communications giant Fujitsu has partnered with U.S. semiconductor and AI leader Nvidia to launch a groundbreaking physical AI robotics initiative, bringing together Japan’s top industrial robotics manufacturers to combine the nation’s legendary manufacturing precision with cutting-edge artificial intelligence technology.

Defined as a new category of intelligent automation, physical AI powers next-generation robots capable of independent, real-time decision-making, rather than being limited to rigid pre-programmed instructions. This technology enables these smart machines to work seamlessly and safely alongside human workers across a wide range of settings, from factory floors and residential homes to medical and care facilities.

The partnership was formally introduced by Jensen Huang, CEO of Nvidia, and Takahito Tokita, CEO of Fujitsu, alongside chief executives from Japan’s three biggest industrial robot makers: Fanuc Corp., Yaskawa Electric Corp. and Kawasaki Heavy Industries. The new collaboration builds on a existing strategic partnership Nvidia and Fujitsu established last year, deepening their shared commitment to advancing AI-powered robotics in Japan.

Executives at the announcement framed the initiative as a targeted solution to one of Japan’s most pressing socioeconomic challenges: an acute and growing national labor shortage, driven by the country’s status as one of the fastest-aging developed nations in the world. Beyond manufacturing roles, the leaders noted that physical AI-powered robots could fill critical gaps in elder care, supporting the growing population of older adults living alone across the country.

Huang emphasized that physical AI is uniquely positioned to leverage Japan’s global reputation for manufacturing excellence. Autonomous moving robots carry inherent safety risks, he explained, requiring the extreme precision and quality control that Japan has long perfected. “Japan’s excellence is a philosophy, a way of life. ‘Made in Japan’ means the highest quality, the highest precision. Japan sets the standard for the state-of-the-art in modern manufacturing,” Huang said, noting that Japan’s iconic “kaizen” (continuous improvement) manufacturing philosophy aligns perfectly with the iterative development needed for safe, reliable physical AI.

While the participating companies have not announced a specific timeline for when these physical AI robots will become widely available in everyday settings, they confirmed that the first phase of the collaboration will launch later this year. No decision has been made to establish a formal joint venture at this stage, though executives left open the possibility of that structure for future phases of the project.

The initiative comes as Japan works to close a perceived gap in global AI development, trailing front-runners including the United States and China. The current Japanese administration under Prime Minister Sanae Takaichi has prioritized tech catch-up, recently unveiling a national strategy to mobilize more than 370 trillion yen ($2.3 trillion) in combined public and private investment for key technology sectors by 2040 — physical AI, semiconductors and data centers are all core priorities outlined in the plan.

Nvidia, the Silicon Valley-based leader in AI computing hardware and open-source AI infrastructure, has rapidly expanded its strategic partnerships across Japan in recent years. Beyond the new robotics collaboration, the firm already has working ties with major Japanese institutions including leading domestic banks, automotive giant Toyota Motor Corp., video game developer Sega, and national research institute Riken.