China rejects Dario Amodei’s call and calls it a “Cold War” strategy

China has reacted strongly to the call by Dario Amodei, CEO of Anthropic, to slow down the development of the most advanced AI systems. Chinese state media described Amodei’s position as a “Cold War” tactic, while Beijing rejected the idea that the United States and its allies should limit China’s access to advanced AI technologies. The reaction came after Amodei argued that the world needs to slow AI development enough to introduce safety mechanisms, while also warning that the United States cannot afford to lose its technological advantage over China. According to Reuters, the dispute has therefore quickly moved from the field of AI safety into the sphere of geopolitical competition.

Why it matters: The new element is no longer Amodei’s call itself, which has already been discussed, but China’s reaction to it. Beijing sees the combination of slowing AI development in the United States and restricting China’s access to advanced chips and technologies as a possible attempt to preserve the American advantage.

AI companies’ shares plunge: Nvidia, ASML, AMD, Micron and SoftBank are hit

Warnings about the pace of AI development quickly reached financial markets. Shares of companies directly linked to the AI boom fell sharply, with Nvidia, AMD, Micron, ASML and SoftBank among the most affected. Stock indexes also came under pressure as investors began reassessing how sustainable the current pace of investment in data centers, chips and AI infrastructure really is. According to Reuters, AI-related stocks were hit globally following warnings from industry leaders.

Why it matters: The stock-market reaction shows that the AI safety debate has also become a financial issue. The industry has built investment plans worth hundreds of billions of dollars on the assumption that model development will continue at a very rapid pace. If that pace slows, the entire economy surrounding AI could be affected, from chip manufacturers and data centers to companies that have invested heavily in infrastructure.

Anthropic, OpenAI and Google discuss creating a joint AI safety body

Anthropic, OpenAI and Google have been discussing since July the possibility of creating a joint body aimed at establishing safety standards for frontier AI models. The discussions began with a proposal for the industry to have an independent mechanism for evaluating highly advanced systems before they are released or expanded. The information was reported by Quartz, amid an increasingly intense debate over the control of models capable of acting autonomously.

Why it matters: We may be seeing the first forms of joint self-regulation among major AI laboratories. If competitors agree to use the same testing methods and safety thresholds, these could become an industry-wide standard and could later influence legislation.

Microsoft proposes its own rules for AI models developed by the company

Microsoft has published a draft code of conduct for its own AI models, the MAI family, establishing clear limits on their behavior. The models should not resist shutdown or human intervention, create their own objectives, or conceal what they are doing. The document also prohibits uses such as weapons manufacturing, certain cyber operations and large-scale manipulation. Microsoft is now seeking public feedback, with the final version expected to guide the development of the company’s models from 2027 onward. The details were reported by GeekWire

Why it matters: Microsoft is turning the principle that “AI must remain under human control” into a series of concrete rules for its own models. One of the most interesting provisions is that a model must not resist correction, interruption or shutdown. This becomes increasingly important as AI moves from chatbots that answer questions to agents capable of taking actions.

Nvidia and Palantir restrict the use of Anthropic models over data concerns

Nvidia, Palantir and Booz Allen Hamilton are considering restricting or reducing their use of advanced AI models supplied by companies such as Anthropic and OpenAI, amid concerns about how customer data is handled. The information, initially reported by The Information and subsequently picked up by Reuters, points to growing concern among companies working with sensitive commercial or government information.

Why it matters: This is a different problem from the risk of AI becoming too autonomous: who controls the data entered into an AI model? For companies, fears that proprietary information could be stored, analyzed or used in unwanted ways could become one of the main obstacles to AI adoption. If this trend expands, companies may increasingly prefer models installed within their own infrastructure or systems offering much stricter guarantees regarding data isolation.

Christine Lagarde warns that Europe risks losing access to AI and calls for its own infrastructure

European Central Bank President Christine Lagarde has warned that Europe must become a producer of AI technology and develop its own computing capacity. In a speech in Vienna, she said dependence on technologies developed outside Europe creates an unprecedented risk that the region could lose access to essential technologies. Lagarde stressed that AI will become important in areas such as healthcare, finance, transportation and border control, and warned that Europe’s data-center shortfall could increase approximately sixfold. According to Reuters Europe must build its own AI capacity to preserve its autonomy.

Why it matters: The statement comes from the head of the ECB and changes the way AI is viewed: it is no longer just a technology, but strategic economic infrastructure. If European banks, hospitals, companies and public administrations become dependent on models and data centers controlled from the United States or other regions, a geopolitical, commercial or legal change could directly affect Europe’s access to AI. Lagarde estimates that effective AI adoption could generate a productivity gain of up to 4% over the next decade.

ByteDance secures $29.6 billion to expand AI infrastructure outside China

ByteDance, the company that owns TikTok, has secured a $29.6 billion syndicated loan from 28 banks, with the funds intended to support the company’s development and, in particular, the expansion of AI infrastructure and data centers outside China. According to Reuters, the financing is one of the largest dollar-denominated lending operations in Asia this year.

Why it matters: The size of the loan shows how much capital is required for the AI race and, at the same time, how Chinese companies are seeking to expand their infrastructure outside China. For ByteDance, building capacity in other countries may also be important because of U.S. restrictions on China’s access to the most advanced chips.

Germany says Europe cannot afford to stop AI development

Germany has rejected the idea of stopping AI development, saying that this is not a viable option for Europe. The German Ministry for Digital Affairs argues that Europe must continue research and innovation to strengthen its digital sovereignty, while at the same time introducing safety measures for AI systems capable of acting autonomously. Berlin is also calling for the United States and China to be involved in an international discussion on AI governance. According to Reuters, Germany believes development and safety must be pursued simultaneously.

Why it matters: Germany’s position outlines a possible distinct European strategy: Europe does not want either to stop AI or to accept unlimited development. Germany is also explicitly linking safety with digital sovereignty. For Europe, unilateral slowdown could mean losing competitiveness to the United States and China, while development without safeguards could increase the risks posed by autonomous systems.

Nine EU countries once again call for a delay to the EES system

France, Germany, Italy, Belgium, the Netherlands, Greece, Malta, Portugal and Switzerland have received additional flexibility for the full implementation of the European Entry/Exit System (EES), which involves biometric registration of travelers from outside the EU, including fingerprints and facial recognition. The system was supposed to become fully operational after a lengthy implementation period, but several countries have cited technical problems, insufficient infrastructure and the risk of long queues at borders. The situation was reported by The Guardian.

Why it matters: EES is a very concrete example of the difference between adopting technology at European level and actually being able to implement it across hundreds of border points. The system must operate at airports, ports and land crossings without disrupting traffic. The fact that nine countries are still struggling to fully implement it shows how difficult large-scale digital infrastructure can be — the same problem now emerging in discussions about data centers and AI infrastructure.