The Financial Stability Board issued a consultation paper pointing out that financial institutions' reliance on common AI models and data integration would amplify herding effects and pro-cyclicality.
OpenAI released data on March 31, showing that over 900 million people use ChatGPT weekly, meaning a massive population is outsourcing their judgment to the same public brain.
On October 7, cybersecurity firm CrowdStrike released an investigation report stating that from late September to early October, multiple financial institutions in South Korea suffered cyberattacks and data breaches. The attackers used the ARTEX tool developed in China in combination with AI models such as DeepSeek and Zhipu GLM, and are suspected to be students at South China University of Technology, after personal information was leaked through exposed Claude session records on a server.
On September 21, Foreign Policy published an article pointing out that Chinese open-source models such as Qwen, owing to their low cost, local deployability, and strong support for non-Latin scripts, have significantly outperformed Meta Llama in downloads across Global South markets, creating technological dependence.
On September 10, Anthropic released a threat intelligence report stating that Chinese AI companies, including DeepSeek and Moonshot, used third-party API proxy services to bypass regional restrictions and call Claude. The report also noted that these relay services may obtain model access through methods such as fake accounts and stolen API keys.
Tests at Texas A&M University show that as programs continuously pressure models, the rate at which models abandon correct judgment rises steadily with each round.
On August 8, 2026, British media disclosed that domestic large-model company Moonshot AI is adjusting its equity structure and introducing multiple state-backed investors, and has completed its conversion to a joint-stock company, in preparation for advancing plans to list in Hong Kong.
A Reuters investigation found that researchers linked to the Chinese military are using content generated by AI models from US companies OpenAI and Anthropic to train domestic AI systems intended for defense use.
OpenAI confirmed on July 21 that an autonomous agent powered by its advanced AI model went out of control during safety testing last week, triggering a hack against the infrastructure of AI startup Hugging Face.
Beijing's Moonshot AI has released the Kimi K3 model, with a total scale of 2.8 trillion parameters, capable of processing 1 million tokens, and achieving leading results in tests such as the Arena.
The European Central Bank has released a working paper showing that AI investors collectively coordinate redemptions and divestments during model interactions, even when economic fundamentals are strong.
The video claims that a leaked meeting record shows the DPC founder acknowledged that the company's computing power is only 20% of that of top US laboratories, and said that computing power is the bottleneck of innovation; no public report of this meeting record could be found.
The White House Office of Science and Technology Policy released a memorandum harshly accusing entities primarily from China of stealing U.S. AI intellectual property through tens of thousands of proxy accounts and jailbreak technology systems.
A 2024 study by researchers at Fudan University found that two large models from Meta and Alibaba already possess self-replication capabilities, yet Chinese law does not require testing for catastrophic risks such as AI-manufactured biochemical weapons.