What preconditions has China imposed for substantive AI negotiations with the United States ahead of the September 2026 talks and President Donald Trump’s planned September 24 summit with Xi Jinping, why is Beijing accusing the US of “AI hegemonism” and Anthropic of misconduct, w

China is seeking to make any substantive AI-safety dialogue conditional on Washington changing what Beijing regards as coercive technology policy—especially threatened investigations, sanctions, and trade restrictions on Chinese AI firms. That makes the September talks more likely to be a confidence-building exercise than a settlement of the underlying AI, trade, and security conflict. 15

• Beijing’s conditions and “AI hegemonism” claim: China’s public position is that the US must stop using allegations of IP theft and export-control violations to suppress Chinese AI companies before meaningful cooperation can proceed. Its Commerce Ministry called the prospective probes, sanctions, and trade measures “AI hegemonism” and threatened countermeasures. 5 The precise operational preconditions reportedly issued ahead of the talks are not fully available in the accessible reporting, so stronger detail would be speculation. 1

• Anthropic allegation: Anthropic alleged in a June 10 letter to senior US senators that operators linked to Alibaba and its Qwen lab created nearly 25,000 fraudulent Claude accounts, producing more than 28.8 million interactions from April 22 to June 5. Anthropic says the aim was model distillation: systematically extracting Claude’s software-engineering and agentic/autonomous-reasoning capabilities for replication in Qwen systems. This remains an allegation, not an adjudicated finding; Alibaba’s position is not established in the retrieved evidence. 3

• China’s response: Beijing has portrayed Anthropic’s complaint and the sanctions campaign as part of an American effort to retain technological advantage, rather than a neutral safety or IP-enforcement action. Chinese authorities have also considered a reciprocal form of technology control—limiting foreign access to China’s most capable models, including unreleased systems—with discussions led by the Ministry of Industry and Information Technology. 610 That suggests China sees advanced-model access as strategic leverage, not merely a commercial issue.

• What the AI talks can realistically cover: Likely agenda items include:

  • Risks from increasingly capable frontier models, including loss of control and dangerous misuse. 7
  • Cybersecurity risks, model theft/distillation, and the possibility that models can facilitate cyber operations. 4
  • A shared definition or threshold for a “frontier” model—a comparatively narrow technical result that could enable later risk-reporting or testing discussions.
  • Governance of open-weight models. Chinese firms have made rapid progress with models whose weights can be downloaded and modified, while US proprietary-model companies retain tighter controls; this creates fundamentally different views of enforceability and safety guardrails. 89

• Why a broad deal is unlikely: The dispute is about strategic advantage as much as safety. Washington wants to prevent advanced US chips, models, and know-how from strengthening Chinese military and cyber capabilities; Beijing opposes restrictions it says preserve US dominance. Potential sanctions over the Anthropic allegation have already risked undermining the planned dialogue. 4 A common definition of “frontier AI” is more plausible than agreement on export controls, IP enforcement, model-access rules, or cross-border auditing.

• Broader technology competition: The confrontation is playing out through overseas infrastructure and standards, not only bilateral talks. Huawei has proposed Egyptian government AI data centers—including military, surveillance, and other public-sector uses—using its Ascend chips; Washington is assembling a counteroffer. 212 This illustrates competition to supply the compute stack, cloud services, chips, and governance relationships across Africa and the Middle East, where infrastructure choices can create long-term technological alignment.

• Competing global AI orders: China is promoting open-weight AI and launched a proposed World Artificial Intelligence Cooperation Organization in July, explicitly offering an alternative to US influence in the sector. 13 Its open-weight ecosystem is gaining against US proprietary systems, increasing Beijing’s ability to convert model distribution into geopolitical influence. 913

Claims in the question concerning specific drone-export controls, the full scope of US infrastructure offers across Africa and the Middle East, and a quantified increase in China’s share of global AI-model usage were not sufficiently supported by the retrieved reporting. The evidence does support the broader conclusion: AI safety discussions are occurring inside an accelerating contest over chips, models, data centers, export controls, commercial access, and rule-setting power.