Donald Trump and Xi Jinping are set to meet in Washington on Wednesday with artificial intelligence prominent on the agenda. The timing follows a global spike in AI safety anxiety, yet both leaders have spent the preceding days signalling that slowing development is off the table. Trump frames AI as a zero-sum race: "Whoever wins AI, wins." Beijing calls slowdown proposals a Western gambit to lock in American companies' lead. The meeting's most concrete output so far is a nuclear-style notification mechanism for AI incidents, discussed between Treasury Secretary Scott Bessent and Chinese counterpart He Lifeng. Bessent called the preliminary talks "very successful" and announced a follow-up meeting in Shenzhen in two months. The ambition is shared threat awareness, not shared regulation — a significant distinction. Domestically, the US legislative path is blocked. House Speaker Mike Johnson dismissed Congress for a seven-week midterm recess, freezing any AI safety legislation. Senator Bernie Sanders has promised a bill prohibiting extremely capable AI, but it has no realistic path through both chambers. Representative Yassamin Ansari called the recess "beyond the pale," warning the technology will continue evolving unchecked during the break. The AI industry itself is split. Anthropic's Dario Amodei and OpenAI's Sam Altman have called for restrictions and nearly reached a joint legally binding safety testing agreement. Nvidia's Jensen Huang and Trump adviser David Sacks push acceleration. Huang stated flatly: "2030 is not going to be the end of the world. There is 0% chance." This internal division within Silicon Valley weakens the regulatory lobby considerably. Europe, which passed the EU's AI Act partly in response to Geoffrey Hinton's 2023 resignation from Google, finds itself sidelined. Ursula von der Leyen's invitation to frontier labs for development pacing discussions drew little response. The continent has regulatory ambition but no frontier models to bargain with. Yoshua Bengio, another AI pioneer, told the Guardian he sees a Covid-style pivot moment approaching, buoyed by Canadian and German governments committing up to £160m for his non-profit focused on "honest AI" guardrails. The funding is real; the optimism may be premature. £160m is a rounding error against the hundreds of billions flowing into frontier model development annually. The structural problem remains unchanged from 2023: safety is a collective action problem, and the two players who must cooperate see each other as the primary reason not to. A notification system is better than nothing. It is not regulation.
