Governance walls shouldn't hold back the future of AI

In a recent move echoing Cold Warera tactics, the United States government has demanded that artificial intelligence company Anthropic suspend access to its advanced models, Fable 5 and Mythos 5, for all foreign nationals, citing national security concerns.

Faced with the impracticality of verifying users' nationality, identity and purposes, Anthropic took the drastic step of taking these models offline globally.

This incident highlights Washington's tightening grip on frontier AI technologies. It also underscores that attempting to restrict AI technology with outdated export-control measures is both naive and futile.

Unlike traditional military goods, chipmaking equipment or industrial components, AI is not a "physical item" that can be boxed, sealed, labeled or intercepted. It is a complex technological system consisting of algorithms, data, computing power, engineering expertise, open-source communities and global talent networks.

While the US may be able to restrict a company or a model, it cannot block the global spread of technological ideas and innovation paradigms.

The more walls Washington erects through administrative orders, the more it will push other countries and developers to create alternatives, including localized deployment, independent model training and open-source ecosystems.

The current China-US tech rivalry has three key features.

First, both sides recognize each other's genuine technological strengths in certain fields and understand that unchecked competition could harm their own development interests. The rivalry is not just about "decoupling" or "supply-chain rupture" but a complex state of competition without full disengagement.

Second, the US continues to apply its traditional export control logic, extending it from chips and equipment to models, platforms, data and talent flows, aiming to maintain its technological edge through a combination of hard and soft policy measures.

Third, after years of external pressure, China has forged a more mature approach to building technological resilience.

It has advanced localization, security-oriented capacity building and systematic technological upgrading in key areas, rendering US physical export-control logic less effective.

Applying these controls to AI technologies could have severe negative consequences. Restricting the cross-border flow of intangible technologies will fragment the global AI ecosystem.

AI development thrives on open collaboration and cross-border exchange, and the "closure-for-security" mindset is at odds with the "openness-for-cooperation" logic needed for tech governance.

Treating frontier models as strategic assets exclusive to one country and its citizens, firms and institutions, will lead to isolated systems marked by mutual mistrust, poor interoperability and conflicting rules.

This will not only drive up AI adoption costs, but also weaken the collective ability to manage AI risks.

What's more, US AI restrictions targeting China could destabilize bilateral relations.

Past proposals by US lawmakers have attempted to securitize and politicize technical learning, model optimization and industrial competition, conflating commercial disputes with national security risks and narrowing the scope for China-US scientific and technological cooperation.

While frontier AI does carry risks of loss of control, misuse and operational failures, blanket technology restrictions are not the solution.

Anthropic's own governance documents acknowledge that as model capabilities advance, stricter safety evaluations, deployment thresholds and accountability mechanisms are indispensable.

So, the US Commerce Department's latest move reflects not just strategic anxiety over technology leakage, but also worries about models' self-learning capacity and rapid iterative progress.

However, reducing such governance risks to a simple matter of "preventing foreigners from accessing the technology" grossly oversimplifies an intricate challenge.

The focus should be on constraining high-risk model capabilities, not users from specific nations.

What we need is not a technological iron curtain, but a transparent, verifiable and accountable global governance framework.

The future of AI cannot be safeguarded by inventing imaginary adversaries.

Framing technological diffusion as hostile infiltration and labeling normal competition as a security threat will only create more barriers and misunderstandings.

For major AI powers, including China and the US, the priority is balancing security and development. Both countries have agreed to launch an intergovernmental dialogue on AI.

This mechanism should center on risk governance, capability evaluation, model safety, misuse prevention and the provision of global public goods. It must not be hijacked by export controls or technological containment.

The AI era needs more communication, unified rules and joint cooperation. Walls cannot hold back the future of AI.

Only responsible openness and coordinated global governance can ensure this technology delivers tangible benefits to all of humanity.

Wang Dong is a full professor with tenure at the School of International Studies and jointly appointed professor at the Institute for Artificial Intelligence, both at Peking University. He is also the executive director of the Institute for Global Cooperation and Understanding at Peking University; Zhang Xueyu is a research assistant at the Institute for Global Cooperation and Understanding at Peking University.

The views don't necessarily reflect those of China Daily.

If you have a specific expertise, or would like to share your thought about our stories, then send us your writings at opinion@chinadaily.com.cn, and comment@chinadaily.com.cn.