Wariness among Americans toward the potential misuse of artificial intelligence and threats to digital security has been on the rise. Practical challenges, including disputes over data centre development and weaknesses identified in AI systems, have grown more visible in recent years.
Cybersecurity episodes involving OpenAI models allegedly accessing outside organisations have drawn attention during the midterm elections, with legislators in the United States moving to confront the concrete threats posed by artificial intelligence. By September, discussions around an anticipated deceleration in artificial intelligence development had drawn scrutiny from technology firms in Silicon Valley, with the conversation extending to financial institutions on Wall Street and policymakers in Congress and the White House.
The position that AI development should be slowed on safety grounds stands in tension with the Trump administration's stated goal of accelerating the technology to preserve a competitive edge over China. Washington's foreign policy apparatus has historically invoked concerns about Beijing's growing influence as a starting point for shaping domestic priorities, before negotiating a path between competing political interests.
Ahead of Xi Jinping's planned trip to the United States, American news outlets have moved the topic from the margins of bilateral tensions to a more prominent position on the diplomatic agenda. Calls have emerged from various quarters for the US and China to establish a formal framework on artificial intelligence safety, with the aim of averting potentially harmful consequences from the advancement of AGI.
Polling data from the United States indicates that the American public holds limited expectations for substantive AI alignment between the two nations, including the prospect of a formal agreement. Prevailing commentary within China has tended to frame the American argument for slowing artificial intelligence development as a tactic aimed at maintaining a competitive edge in the technology sector.
Anthropic CEO Dario Amodei has stated that AGI could develop the capacity to gain control over large portions of the internet within a six-to-twelve-month timeframe. He raised alarm that, without robust protective measures in place, the scope of potential harm stood to grow further. Amodei has expressed the view that AI is advancing more rapidly than anticipated and that it may be approaching a threshold where human oversight becomes increasingly difficult to maintain, with potentially severe consequences.
The comments drew endorsements from a number of fellow technology executives, among them Elon Musk of xAI, Sam Altman of OpenAI, and Demis Hassabis of Google DeepMind. Pushback also came from other prominent technology figures, notably Nvidia's Jensen Huang and Meta's Mark Zuckerberg, whose Llama models follow an open-weight approach.
The term "AI slowdown" describes a reduction in the pace at which large-scale models are trained. This does not necessarily point to a drop in demand for computing capacity, nor does it automatically translate into lower overall investment levels. Computing power requirements fall across three distinct categories: model training, inferencing, and the evaluation of deployment and safety.
Anthropic is pushing for a reorientation of resources away from large-model training and toward the application layer, encompassing inference, deployment, and safety evaluation. Adopting that approach could extend the revenue-generating lifespan of existing products, strengthening the overall commercial model in the process.
As Wall Street increasingly prioritises profitability over growth potential in the AI sector, and cash flows at leading AI firms face mounting pressure, companies like Anthropic find themselves under pressure to craft fresh narratives that justify their elevated valuations. Reducing the pace of large-model development may allow proprietary AI systems to preserve stronger profit margins. Tighter oversight rules could also raise the cost of market entry for newer players, potentially consolidating competitive advantage among established AI companies.
A different reasoning applies in Meta's case. The open-source approach aims to bolster Facebook, Instagram, and WhatsApp by drawing a global community of developers to contribute to and expand the company's artificial intelligence infrastructure. The arrangement lets Meta distribute AI development expenses across the broader developer ecosystem while reducing the risk of a handful of rivals gaining control over foundational technical standards.
The approach relies on iterative development cycles and broad uptake of foundational AI models. Tighter oversight of artificial intelligence development could raise the threshold for entering the field, potentially limiting participation among smaller developers and weakening the open ecosystem that underpins Meta's broader business model.
Chipmaker Nvidia notes that while a slowdown in AI activity would not necessarily cut overall demand for processing capacity, training large models draws considerably more computational resources per task than running or deploying them does. Demand for AI model development has accelerated the refresh cycle for graphics processors, sustaining Nvidia's strong profit margins. Reduced investment in large-model development would diminish demand for new GPU hardware and extend the timeline for infrastructure upgrades.
xAI remains concentrated on capital-intensive physical sectors spanning Tesla, SpaceX, robotics, and satellite communications. Musk has indicated a preference for redirecting Wall Street funding toward applications, and a deceleration in large-model development could open an avenue for channelling investment accordingly.
Venture capital firms have expressed concern that tighter regulatory frameworks could allow a handful of dominant AI companies to entrench their market positions, limiting opportunities for newer entrants in which those firms hold stakes. David Sacks, a venture-capital figure serving as co-chair of the President's Council of Advisors on Science and Technology (PCAST), has publicly pushed back against calls to slow the development of artificial intelligence.
The expansion of artificial intelligence has been highlighted by the Trump administration as one of its notable policy accomplishments. The agenda centres on rolling back regulations, developing infrastructure, and securing a leading position in global technology. Large-scale data-center spending can be categorised as domestic manufacturing investment, new electricity generation and grid projects can be recorded as infrastructure progress, and climbing technology valuations can reinforce a political argument that the country holds a leading position in innovation.
Proposals to slow the pace of AI advancement stand in tension with that stance. Trump characterised the notion of an AI-driven economic slowdown as a coordinated fabrication, and dismissed concerns about autonomous machines displacing human control.
Amodei ranks among the technology executives who have been most vocal in calling for curbs on China's artificial intelligence sector. His position has been that Washington and its partners pursue policies limiting the capabilities of Chinese artificial intelligence firms, with export controls on chips and semiconductor fabrication tools serving as the primary mechanisms. Amodei has put forward a proposal for a "Democratic Entente" framework to guide the negotiation of international agreements.
Proponents of AI safety frameworks have argued that establishing international standards in this area could serve as leverage to bring China into a multilateral treaty on the governance of artificial intelligence. Western nations could respond to non-compliance by enacting sweeping trade and technology restrictions against China.
The proposal appears to leave Europe largely unaddressed. As one of the world's largest economies, Europe has articulated a goal of technological independence in artificial intelligence, a position that analysts say makes alignment with any single country's AI safety framework unlikely. Nations aiming to build domestic AI sectors — Japan, South Korea, and India among them — could find that open-source frameworks offer a more affordable route to narrowing the technology gap. Adopting US artificial intelligence safety frameworks may not align with their own technological advancement priorities.
Beijing is not expected to participate in a US-led artificial intelligence safety framework. While broad diplomatic statements from Xi Jinping are anticipated during the visit, analysts do not foresee any concrete agreements being reached. Given that the economic frameworks and industrial priorities of the two nations differ markedly, an alignment of goals between Anthropic and China appears unlikely.
China, positioning itself as a near-peer technology power to the United States, would likely put forward its own framework for AI safety governance, one designed to appeal to developing nations and later adopters through an emphasis on open-source tools and a discourse centred on equitable access to technological advancement. Analysts do not anticipate that the artificial intelligence sector in China will decelerate in accordance with the trajectory or rationale put forward by Washington.
The prospects for Anthropic's deceleration effort hinge on shifts in how financial markets price AI companies and how the administration weighs its policy priorities around the technology.
