In an AI crisis, would China answer the phone?

After the September 24th Xi–Trump meeting, the Trump administration says the two countries will set up a channel for AI-related incidents and revive an AI dialogue before a November leaders’ meeting. Cold-war hotline analogies meet a sceptical history: China refused a Pentagon call after the 2023 spy balloon; a 1998 presidential hotline sat unused for a week after the Belgrade embassy bombing; after the 2001 spy-plane collision, US officials struggled for 12 hours to reach Beijing; a 2008 military hotline required 48 hours’ notice and went unanswered in a 2009 South China Sea clash. Daniel Kritenbrink and other veterans call past channels mostly ineffective—China often goes dark, delays for leverage, or treats US “guardrails” as containment. Shared interest in rogue models and malicious non-state actors could still create common ground. Drew Thompson argues a workable AI channel should be text-based for committee review rather than live voice; thresholds, AI-expert staffing and regular meetings modelled on the Military Maritime Consultative Agreement could help. Details remain unclear, including who sits in the Bessent–He Lifeng dialogue, while trade talks compete for attention.