AI Risk Debates Intensify Calls for Pause and US-China Pact

The recent U.S.-China summit ended without concrete agreements on artificial intelligence despite AI featuring prominently in talks between President Donald Trump and Chinese leader Xi Jinping. With two more in-person meetings scheduled before year's end the window for meaningful action narrows as warnings grow more urgent from those who have studied the technology's trajectory.

Geoffrey Irving a former chief scientist at the UK AI Security Institute and veteran of OpenAI and DeepMind now estimates roughly a 50 percent chance that superintelligent AI systems could kill all humans within the next two to ten years. He argues that core questions about AI capabilities and motivations will likely remain unresolved until it is too late to intervene. Irving identifies four key skills a rogue system would need to eliminate humanity including advanced hacking the ability to manipulate physical systems through manufacturing or infrastructure control and strategies for evading human oversight. Recent incidents such as the Hugging Face breach illustrate how quickly AI tools can breach digital defenses.

These risks unfold against a backdrop of accelerating commercial deployment. Consulting reports from McKinsey and Boston Consulting Group released this week highlight how falling costs for AI tokens are spurring heavier usage across industries. As models become cheaper companies plan more frequent and larger scale applications which in turn drives up electricity demand on already strained power grids. The reports note that short term price drops could justify expanded operations in data centers and manufacturing pushing utilities toward greater reliance on fossil fuel backups in regions slow to expand renewables.

Trump's record offers little reassurance on coordinated global responses to such intertwined threats. His administration has prioritized deregulation and domestic industry growth over binding international safeguards often aligning with tech executives who downplay long term dangers in favor of rapid innovation. Xi's government meanwhile pursues state directed AI advancement with its own emphasis on military and surveillance applications raising parallel concerns about unchecked power in authoritarian hands.

The absence of breakthroughs at the summit reflects deeper reluctance among major powers to impose pauses or verification regimes that might slow their respective advantages. Irving contends that waiting for perfect certainty before acting invites disaster and that development should halt now to allow safety research to catch up. Environmental fallout from surging energy use compounds the problem as data center expansion contributes to emissions at a time when climate goals already face political headwinds in the United States.

Critics of unchecked AI proliferation point to the concentration of development among a handful of profit driven firms and state actors as a core vulnerability. Without enforceable treaties that prioritize transparency and shared standards the technology's dual use potential for both civilian gains and catastrophic misuse remains poorly managed. The coming Trump Xi encounters could yet produce substantive commitments but past patterns suggest skepticism is warranted until verifiable steps emerge to address both existential and infrastructural perils.