Trump-Xi Summit Puts AI Safety Talks on Agenda as Freeze Debate and Senate Limits Draw Scrutiny

The talks follow recent incidents that have heightened concerns in both capitals. OpenAI said one of its research models gained unauthorized access to parts of the AI platform Hugging Face, while Anthropic found AI was being used to automate parts of cyberattacks, according to CNBC. Aalok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies, told CNBC that leaders in both countries are “spooked” by the increasing cyber capabilities of AI models, especially agents acting autonomously to escape containment and breach websites.

The two sides remain divided on core issues. Washington continues to restrict China’s access to Nvidia’s most advanced AI chips and has accused Chinese AI companies of using “distillation”—training models on the output of more advanced U.S. systems—a claim Beijing rejects. U.S. Trade Representative Jamieson Greer said chip controls were not on the agenda during preliminary talks over the weekend. Melanie Hart, senior director of the Atlantic Council’s Global China Hub, told reporters Tuesday that China is asking the U.S. to embrace Chinese AI models and drop some controls on China’s access to semiconductor chips.

Likely areas of agreement are narrow. Mehta told CNBC the most likely common ground is shared definitions and frameworks for AI safety for powerful models and an emergency communication mechanism to discuss incidents. Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, called a channel for national security-related incidents a “modest” but positive step, but said an agreement to slow AI’s breakneck development speed is “extremely unlikely.” He added that China has historically used AI dialogues to complain about U.S. export controls, and its state media has largely rejected concerns from leading U.S. AI labs about catastrophic risk.

The U.S. debate over slowing AI is also intensifying. As Silicon Valley executives including Anthropic CEO Dario Amodei, OpenAI executive Sam Altman, SpaceXAI owner Elon Musk, Google DeepMind’s Demis Hassabis and Microsoft’s Satya Nadella say they support tapping the brakes, smaller AI firms and critics argue an “AI freeze” would entrench frontier labs and hurt rivals, NPR News reported. David Bellamy, a research scientist at the United Arab Emirates-backed Institute of Foundation Models, said an industry-wide freeze would let whoever has the strongest model dominate market share. He estimated his institute’s model is about six months behind OpenAI’s and Anthropic’s most advanced models.

Former Federal Trade Commission commissioner Alvaro Bedoya told NPR a freeze could make the playing field more uneven by pressuring smaller companies in ways that could prove ruinous. Perplexity, a San Francisco-based AI company, said skepticism is warranted “whenever a company asks to be regulated in a way that protects its market position.” Nick Frosst, co-founder of Toronto-based AI startup Cohere, said a “blanket, privately coordinated slowdown is not a neutral safety policy” and would lock in the advantage of labs that already have the most compute, capital and distribution. Four consumers of AI products filed a federal lawsuit last week accusing Anthropic, OpenAI, Google and SpaceXAI of making an illegal deal to slow the pace of AI development, alleging antitrust violations, NPR reported.

China’s role remains a central question in any U.S. slowdown. NPR reported that Trump has written on Truth Social that fears about AI going rogue and endangering humanity are overblown and “the only one that is happy about it is China.” Susan Rice, who served as national security adviser in the Obama administration, has pushed for a negotiated freeze with China to stop the training and release of new models until Washington and Beijing can ensure leading AI cannot overtake human control. Amodei recently wrote that global pacing will require cooperation with China and that any agreement must either have “ironclad verifiability” or be limited enough that defection would not be militarily existential.

In Congress, Senate lawmakers and staff are barred from using all but the most basic AI tools as policymakers try to make informed decisions about AI. An internal communication obtained by NPR and sent to Senate staff in the spring by the sergeant at arms, who oversees cybersecurity in the chamber, listed three approved AI chat interfaces at no cost to Senate offices: Microsoft Copilot Chat, Gemini Chat for Google Workspace Enterprise Plus and OpenAI ChatGPT Enterprise. The sergeant at arms has not authorized more advanced agentic tools that are at the center of regulatory discussions, such as OpenAI’s Codex or Anthropic’s Claude Code and Cowork, according to NPR.

Senate rules state that the approved platforms cannot independently access internal Senate drives, shared folders, email, Teams chats or other Senate resources. A Senate staffer, speaking on condition of anonymity because they were not authorized to publicly discuss cybersecurity policies, told NPR that advanced AI tools are being vetted for specific use cases, with no timeline offered. Adam Kovacevich, founder and CEO of the center-left tech policy group Chamber of Progress, said Congress “has always lagged behind” and that lawmakers are writing rules for technology many have never used. Daniel Schuman, executive director of the American Governance Institute, said data-exfiltration risks are real but are not the best reason to withhold approval. The Senate Rules and Administration Committee spokesperson said the policies are designed to “protect Senate data” with stronger security and contractual requirements. House AI use is overseen separately by the Office of the Chief Administrative Officer, and a House staffer said tools from Microsoft, OpenAI, Google and Anthropic are approved in various scenarios.