Introduction
In recent years, the global discourse on artificial intelligence (AI) has reached a critical juncture. High-profile resignations, corporate manifestos, and escalating warnings from American AI executives have called for deliberate slowdowns in frontier AI development to prevent existential misalignments. However, political leaders, state media, tech executives, and academic researchers in China have expressed deep skepticism toward these Western AI safety narratives.
Chronology of Key Events
- November 2023: China joins the U.S., EU, and 26 other nations at the UK AI Safety Summit, signing the Bletchley Declaration and endorsing international cooperation on AI risk management.
- May 2024: U.S. and Chinese officials hold high-level bilateral talks in Geneva on AI risks, highlighting divergent views on military applications and international governance frameworks.
- February 2026: Reports surface that domestic pressure on Chinese AI startups to close the performance gap with U.S. rivals leads to resource concentration on raw model capabilities over dedicated domestic safety research.
- September 10, 2026: High-profile resignation of Anthropic safety researcher Jacob Coxon over existential risk (x-risk) concerns sparks renewed global calls for AI development pauses.
- September 12, 2026: Anthropic CEO Dario Amodei publishes "We Must Pace the Frontier," advocating industry-wide capability slowdowns while explicitly urging strict U.S. semiconductor export enforcement and crackdowns on foreign "model distillation."
- September 14, 2026: China's Cybersecurity Standardisation Technical Committee (TC260) releases the AI Safety Governance Framework 3.0, introducing national guidance on autonomous AI agent risks, open-source safety, and supply chain security.
- September 15, 2026: Chinese Foreign Ministry spokesperson Guo Jiakun publicly rejects Western "threat narratives," warning against fearmongering and stating that AI regulation should not be weaponized to restrict developing nations.
- September 18, 2026: Chinese state media (Global Times) characterizes Western safety pacing proposals as a "Cold War playbook" containing provisions designed to entrench U.S. tech monopolies.
- September 23-25, 2026: U.S. President Donald Trump and Chinese President Xi Jinping hold a state summit in Washington, D.C. The two nations agree to establish formal bilateral AI communication channels and notification mechanisms regarding high-level risks, while maintaining competing views on global governance.
Key Findings: The Core Reasons for China's Skepticism
1. Perception of Safety Calls as Geopolitical Containment ("Ladder-Pulling")
- Weaponized Regulation: Chinese industry insiders and political commentators frequently describe Western AI safety proposals as an attempt at "ladder-pulling"—using safety rhetoric to lock in existing U.S. advantages.
- Targeting Distillation and Compute: In Dario Amodei's September 2026 proposal, calls for voluntary industry slowdowns were combined with policy recommendations to tighten U.S. chip export controls and restrict "unauthorized model distillation" by foreign competitors.
- Protecting Market Moats: Analysts note that open-weight and cost-efficient models from Chinese firms (e.g., DeepSeek, Alibaba's Qwen, Moonshot AI) have challenged the economic moats of Silicon Valley's closed-source models. Consequently, Beijing views Western calls for mandatory pre-deployment safety pauses as a ploy to slow down agile Chinese competitors.
2. Divergence in Risk Priorities: Existential Doomsday vs. Political and Systemic Risks
- Speculative vs. Concrete Risks: Western AI safety discourse heavily emphasizes "existential risk" (x-risk)—such as rogue general intelligence causing human extinction or escaping digital containment. Chinese scholars (e.g., Kyle Chan of the Brookings Institution) point out that Chinese leadership views x-risk scenarios as speculative or culturally alien.
- Focus on Regime Security and Content Control: Beijing defines AI risk primarily in terms of political stability, regime security, and social cohesion. China's State Security Minister Chen Yixin warned in September 2026 that AI poses immediate risks of "cognitive warfare," deepfake propaganda, and social destabilization.
- Strict Domestic Rules with Different Intent: While Western observers claim China ignores safety, domestic scholars note China has implemented strict regulations regarding algorithmic recommendations, deepfake labeling, and compliance with "socialist core values." China regulates for immediate state control and content safety rather than long-term superintelligence alignment.
3. The Credibility and Verification Paradox
- Lack of Observability: Chinese AI executives and researchers ask how mutual restraint could ever be enforced. As noted in a September 2026 ThinkChina analysis, Chinese labs have no reliable, observable way to verify whether American frontier labs (such as OpenAI, Anthropic, or defense-funded projects) have actually slowed down their training runs.
- Fear of Unilateral Disarmament: Without verifiable inspection regimes, Chinese decision-makers view agreeing to a domestic AI slowdown as unilateral technological disarmament in an active superpower race.
4. Multilateral Governance Hegemony vs. "Equal Footing"
- Rejection of Western Coalitions: Beijing rejects Western-led regulatory bodies or "democracies-only" AI safety coalitions that attempt to impose rules on global markets.
- Demand for UN Centeredness: In line with statements from China's Foreign Ministry in September 2026, China insists that global AI governance must occur on an "equal footing" through the United Nations. Beijing positions itself as a champion of the "Global South," arguing that Western-led safety rules threaten to create an AI divide that deprives developing nations of access to beneficial technologies.
5. Resource Constraints and Competitive Pragmatism
- Compute Shortages Force Capability Focus: U.S. sanctions have created an acute compute deficit for Chinese developers. Former DeepSeek researcher Tu Jinhao stated in February 2026 that Chinese firms allocate nearly all available computational and capital resources to closing the capability gap with the U.S., leaving minimal compute budgets for dedicated safety alignment teams.
Key Stakeholder Positions and Disputed Claims
| Stakeholder / Entity | Primary Stance / Claims | Key Source & Date |
|---|---|---|
| Dario Amodei (Anthropic CEO) | Advocates "pacing the frontier" via voluntary lab slowdowns, third-party embedded evaluators, stricter U.S. chip export controls, and bans on model distillation. | Essay "We Must Pace the Frontier" (Sept 12, 2026) |
| Chinese Foreign Ministry (Guo Jiakun) | Rejects Western "threat narratives" and AI fearmongering; calls for equal focus on development and safety, multilateral UN governance, and "equal footing." | Foreign Ministry Press Briefing (Sept 15, 2026) |
| Global Times (Chinese State Media) | Labels Western AI pacing manifestos a "Cold War playbook" designed to maintain Western technological monopolies under the guise of safety. | Publication (Sept 18, 2026) |
| Ministry of State Security (China) | Warns that foreign AI models pose cyber espionage, backdoors, and political stability risks; focuses safety on cognitive warfare and data security. | Official Statement / Report (Sept 14, 2026) |
| Future of Life Institute (Max Tegmark) | Criticizes Chinese firms (e.g., DeepSeek) for a lack of safety transparency as their capabilities approach top Western models. | FLI Report (Dec 2025 / Feb 2026) |
Disputed Claims
- "China is ignoring AI safety."
- Western Claim: Western non-profits and executives argue Chinese labs lack transparent safety protocols and fail to fund safety teams at levels matching their capability gains.
- Chinese Counter: Chinese officials point to domestic regulations (TC260 Governance Framework 3.0, deepfake mandates) to argue China enforces stricter governmental safety rules than the West, but prioritizes content control and application safety over theoretical x-risk.
- "AI pacing benefits all humanity equally."
- Western Claim: Industry leaders argue slowing down capability advances prevents catastrophic loss of control, benefiting all nations equally.
- Chinese Counter: Chinese analysts argue that pacing without global compute parity freezes the current U.S. lead in place, harming latecomers.
What Remains Uncertain
- Operationalization of US-China AI Bilateral Talks: While President Trump and President Xi agreed in September 2026 to launch a formal AI risk notification channel, it remains uncertain whether these talks will yield concrete verification protocols or remain high-level political dialogues.
- Enforcement of TC260 Framework 3.0: China's AI Safety Governance Framework 3.0 released in September 2026 introduces guidelines for autonomous AI agents, but it remains a non-binding reference framework. It is unclear if Beijing will transform these recommendations into statutory mandates for third-party lab audits.
- Impact of Open-Source Models: As Chinese open-weight models (e.g., Qwen, DeepSeek) continue to be adopted globally due to lower costs, the effectiveness of Western closed-source safety agreements without binding Chinese participation remains an unresolved challenge.