China and the US seek protection mechanisms against AI risks

Artificial intelligence is no longer just answering questions and is starting to act. The growth of AI agents capable of using tools, executing commands and making decisions has expanded the usefulness of the technology, but also the possible consequences when it fails or is misused.

In a February 2026 report on international AI security, Turing Award laureate Yoshua Bengio grouped the risks of advanced AI into 3 broad categories: malicious use, malfunction, and systemic risks.

O malicious use includes employing AI to carry out cyber attacks, commit financial fraud and promote manipulation through deepfakesor potentially assist in the development of biological or chemical weapons.

O malfunctionon the other hand, can occur without malicious intent, when models make errors due to capacity limitations, misaligned objectives, or inappropriate interactions.

Os systemic risks arise from the cumulative impact of AI on society, including employment, intellectual property, information ecosystems and the distribution of economic benefits.

Together, these risks present policymakers with the same dilemma: governments cannot wait until major losses occur before responding, but they also cannot turn all uncertainty into a constraint on innovation. The challenge is to protect against serious risks while leaving room for technological development.

China: development and security in parallel

China began defining its approach relatively early. The State Council’s 2017 New Generation Artificial Intelligence Development Plan identified AI as a strategic technology, while warning of risks to employment, privacy, ethics, economic security and social stability. The plan set goals for creating laws, ethical standards and security assessment capabilities in parallel with the development of the sector itself.

Since then, China has developed a layered regulatory system rather than a single comprehensive law on AI. Its Cybersecurity Law, Data Security Law and Personal Information Protection Law form the basis, complemented by rules dealing with algorithmic recommendations, deep synthesis, generative AI and, more recently, anthropomorphic AI services.

This structure reflects, in part, the speed at which technology is changing. Xue Lan, dean of Tsinghua University’s Schwarzman College and head of the institution’s Institute for International AI Governance, said China’s approach is evolving from responsive governance to focused governance and then toward what he calls “agile governance”: responding quickly to emerging risks and adjusting rules as technology and markets evolve.

This approach can be seen in several regulatory mechanisms that have become recurring features of China’s AI framework.

Providers of certain algorithms with the capacity to influence public opinion or mobilize society are required to register them with regulatory bodies. The rules governing deep synthesis and generative AI impose requirements involving data management, content review, security assessments, and labeling of AI-generated material.

Regulations that took effect in July 2026 for anthropomorphic AI services added safeguards aimed at identification, overuse, psychological risks, data security, and privacy.

China has also developed specific rules to identify AI-generated content. Regulations and a mandatory national standard that took effect in 2025 established requirements for visible and embedded labels in text, images, audio, video and virtual content.

Parallel to these specific rules, there is a broader system of ethical analysis. Measures introduced in 2026 established procedures to review AI research and development that may pose risks to human dignity, public order, health, the environment and sustainable development.

Together, these measures produced a governance model based on algorithm registrations, security assessments, content labeling, and risk-based regulation. These measures place the main responsibility on companies, combining regulatory oversight, industry self-regulation and public participation.

There are still limitations. China’s current tools are more geared toward dealing with malicious use and technical failures, while deeper systemic issues — including changes to the labor market and intellectual property regimes — are more difficult to address through specific AI regulations alone. Addressing these issues will require closer integration with areas such as civil, labor and intellectual property law.

USA: innovation first, but not without risks

The administration of President Donald Trump (Republican Party) has taken a markedly different approach at the federal level. In short, its policy prioritizes innovation, reduces regulatory burdens, and seeks to ensure the U.S. wins the global AI race.

Shortly after beginning his 2nd term, Trump signed an executive order to remove barriers to US leadership in the AI ​​sector. His administration then released an AI Action Plan in July 2025, focused on accelerating innovation, expanding infrastructure, and strengthening U.S. leadership abroad.

Later decrees sought to speed up approvals to deploy AI infrastructure, promote exports of the entire suite of U.S. AI technologies, and limit what the government considers excessive state regulation. The administration has repeatedly argued that North American AI companies need room to innovate without a tangle of conflicting rules among the 50 states.

Still, Washington has not abandoned AI safety. A national policy framework for AI from the White House, released in March 2026, addressed child protection, intellectual property, small business, free speech, workforce education, and limits on state regulation, while maintaining the administration’s pro-innovation stance.

In June, Trump signed another executive order establishing a voluntary mechanism for cooperation between advanced model developers and the federal government on cybersecurity risks. The order provides confidential assessments to measure advanced cyber capabilities and allows developers of advanced models to voluntarily grant the government controlled access before making them available to other trusted partners.

The distinction from China is important. China’s framework relies, in part, on mandatory security registration and assessment mechanisms. The US federal approach emphasizes voluntary cooperation and explicitly avoids turning the process into a system of licensing or pre-launch approval.

Meanwhile, state governments have focused more heavily on the risks arising from specific uses of AI.

Several States have introduced rules for chatbotsparticularly when they involve minors and mental health issues. Measures include requiring users to be informed that they are interacting with an AI rather than a person, restricting sexual content and manipulative interactions with minors, introducing parental controls, and establishing protocols for responding to signs of suicide or self-harm.

California and New York went further. The California Advanced Artificial Intelligence Transparency Act requires developers of advanced models to establish security frameworks to address risks of catastrophic proportions and imposes requirements for disclosure, incident reporting, and protection of model weights. New York’s Responsible AI Safety and Education Act takes a similar approach.

The result is a U.S. system with two distinct layers: a federal government heavily focused on innovation and international competitiveness, and states that increasingly regulate AI-specific risks and, in some cases, advanced models themselves.

Space for dialogue

China and the US take different approaches to AI governance — China through a layered system of registrations, labeling, ethical review and agile regulation, and the US through a more innovation-focused framework that combines voluntary federal measures with state rules — but both face the challenge of technology advancing faster than regulation.

Despite intensifying competition for AI leadership, the 2 countries also share concerns about risks such as model proliferation, cyberattacks and misuse by non-state actors, creating space for cooperation.

Previous government-level dialogues and the agreement reached during President Trump’s visit to China in May 2026 to resume dialogue on AI suggest that competition and communication can coexist, indicating that continued dialogue and the adoption of risk safeguards are important for both countries and for global stability.

This report was originally published in English by Caixin Global on August 20, 2026. It was translated and republished by Poder360 under mutual content sharing agreement.