A growing chorus of researchers and industry leaders warns that the U.S.-China race for AI dominance is escalating existential risks, with mutual distrust pushing both nations toward a scenario one analyst says could drive humanity "off a cliff." The September 24 meeting between President Trump and Chinese President Xi Jinping in Washington is now expected to include AI security on the agenda.
The core long-term fear, amplified recently by an Anthropic researcher, is that a future superintelligence - an AI agent smarter than any human - could view people as competitors for resources. Leaders at OpenAI and Anthropic have publicly called for a slowdown to "pace the frontier."
Competing warnings and the call for guardrails
China has issued its own alerts. Yi Zeng, dean of the Beijing Institute of AI Safety and Governance, wrote that "the catastrophic risks posed by AI, and the hidden danger of human extinction, are growing by the day." A separate U.S. research team last week used AI to expose a security flaw in WeChat, the platform used by over 800 million people in China.
Britain's King Charles III is convening industry leaders in Scotland on Thursday to explore shared principles for AI development. Nvidia CEO Jensen Huang, whose firm is attending, said in April that U.S. and Chinese researchers must communicate. "This bilateral coordination by industry leaders is glaringly missing because of our current attitude about China as an adversary," he said.
Yet political rhetoric remains combative. Republican Senator Ted Cruz said last week that "if there are gonna be killer robots, I'd rather they be American killer robots than Chinese killer robots." President Trump dismissed AI threat warnings as a "sick conspiracy" and posted on Truth Social that "WHOEVER WINS AI, WINS!"
Distrust blocks the safety conversation
Leia Wang, non-resident scholar at the Carnegie Endowment for Peace, told CBS News that for U.S. leaders, "the number one concern is technological competition" when discussing AI and China. China's government has rejected calls to slow development, viewing the concept as a U.S. attempt "to assert its technological dominance," Wang said.
After Anthropic CEO Dario Amodei advocated for democracies to maintain an AI lead over autocracies, China's Foreign Ministry responded that "narratives of threat … serve only to disrupt the process of global AI governance." Wang noted that reducing the "AI safety conversation into a technological competition conversation" is "the most unhelpful framing" given the stakes.
National University of Singapore lecturer Alex Capri wrote in February that both countries "face a prisoner's dilemma … both would be safer with restraint, yet each accelerates development to avoid falling victim to the other."
Paths forward without a grand bargain
A coordinated international slowdown may not be the right frame for engaging China, according to Kwan Yee Ng, Head of International Governance at Concordia AI. She pointed to "low-hanging fruit" including sharing cybersecurity best practices, coordinating against misuse by terrorist groups, and opening low-level government talks to exchange technical information.
Matt Sheehan, also at the Carnegie Endowment, proposes "AI safety in parallel," where each country independently reduces risks because it serves their own national security, regardless of what the other does.
For developers and engineers working in Generative AI and LLM fields, these geopolitical fault lines are not abstract. The regulatory frameworks and safety protocols debated today will shape model access, training data restrictions, and cross-border collaboration for years. Professionals building systems on frontier models should track bilateral talks closely - a sudden agreement or breakdown could directly affect tooling availability and compliance requirements.
Why this matters for IT and development professionals
The push-pull between acceleration and safety creates immediate technical consequences. If nations adopt parallel safety tracks, expect divergent compliance standards that complicate deployment across markets. Ng's emphasis on granular technical exchanges suggests future roles for engineers in red-teaming and cybersecurity coordination, not just policy advisors. For those working in AI for IT & Development, building adaptable architectures now - ones that can accommodate sudden regulatory shifts - is a practical hedge against the uncertainty Wang describes.
Your membership also unlocks:
