The Empty Chair at Xi's Banquet That Redefined US AI Diplomacy

One Empty Chair, One Private Table

On a Tuesday evening in late June, Donald Trump sat across from Xi Jinping in the Diplomatic Reception Room of the White House. Beside the Chinese president’s table sat Sam Altman, Jensen Huang, Elon Musk, Mark Zuckerberg, and Jeff Bezos. The American AI industry’s front line, arranged in a row for a photo op that doubled as a diplomatic message: we build together, and we build here.

One seat was empty. Dario Amodei, chief executive of Anthropic, had a scheduling conflict.

Two days later, Axios reported, Trump would host Amodei alone. No entourage, no Chinese dignitaries, no press pool. A single dinner for two men who had been circling each other for months through intermediaries, objections, and a federal agency that had briefly frozen Anthropic’s model exports. It is the first one-on-one between the president and the company’s founder, and the gap between the two events—the public banquet and the private plate—does most of the analytical work for anyone trying to read where American AI policy actually sits in 2025.

The First American Company Slapped With a Foreign-Sanctions Label

To understand why that empty chair mattered, you need the February order that almost broke the administration and the company apart. The Department of Defense added Anthropic to its “supply chain risk enterprise” list, directing every federal agency and military contractor to halt transactions with the lab. The list, up to that point, had been a tool aimed squarely at Chinese and Russian firms. A domestic startup landing on it was not just unusual; it was, by every protocol manual, unprecedented.

The trigger was a clause. The Pentagon wanted Anthropic’s models licensed for “all legal purposes.” Amodei’s team refused. Anthropic’s internal safety charter caps deployment: no fully autonomous weapons, no mass-surveillance architecture aimed at American citizens. That refusal earned the company a designation normally reserved for entities whose code runs on Beijing or Moscow servers. The Commerce Department, in a parallel move, briefly placed export-control directives on Claude model releases, suspending service until the paperwork cleared.

The episode played out like a cold-war containment memo crossed with a Silicon Valley HR dispute. A lab that builds some of the most sophisticated reasoning systems in the country got treated, procedurally, like a hostile foreign actor because it declined to open its back door to a defense acquisition program. The optics did not survive the weekend. Within weeks, White House aides reportedly balked at scheduling Amodei for a bilateral, one official telling colleagues the CEO was “too weird” for the president’s calendar.

The Thaw, and the Document That Tried to Stop It

The thaw came slowly. Amodei’s public letters advocating a pause in frontier-model development drew a sharp jab from Trump, who called the argument “a farce” (the Korean source renders it as a manipulation or staged act). The criticism landed on social media the way Trump’s rhetoric usually does: blunt, personal, aimed at the individual rather than the institution.

So when the one-on-one dinner surfaced as a confirmed item on the 27th, critics of Amodei moved fast. Axios obtained a memo, styled deliberately in the president’s own register of exclamation points and dismissal, that was circulated inside the White House to cool momentum. The document recited Amodei’s “long record of attacking the president,” flagged his ties to Democratic donors, framed the DoD standoff as proof that “Dario does not trust the president,” and capped the argument with a taunt about Amodei’s prediction that AI would eliminate half of new white-collar roles within the year: “Wrong!”

Whether the memo changed the schedule is unclear. The dinner still happened, or was scheduled to happen, as Axios reported it. But the existence of the document tells you the temperature inside the West Wing. This is not a policy disagreement handled in a briefing room. It is a personal feud, fought with the same tools as a campaign attack ad, over whether a CEO gets a seat at the table where the next export rules get drafted.

What the Xi Banquet Ordering Tells You About US-China Tech Logic

The timing is the second signal, and it is subtler. Trump hosted a state dinner for Xi the same week. Five of the six largest tech and AI figures in America were seated as guests of honor. The message to Beijing was legible: the United States fields its AI champions at the diplomatic table, not in back rooms. The message to the domestic industry was different: your CEOs are the president’s dinner guests, which in the Trump calculus means you are in the good graces.

Amodei’s absence from that banquet was, on the surface, a calendar hiccup. In practice, it exposed a fault line. The other five CEOs were there to project unity. Amodei’s relationship with the administration was actively adversarial, and dragging him into a photo op with a Chinese head of state while the DoD had just frozen his company’s exports would have looked like a capitulation to his own critics. Trump, who governs by proximity and personal loyalty, noticed the absence. The invite followed. The one-on-one format was not generosity; it was a way to fold Anthropic back into the inner circle without the spectacle of a joint press conference.

For English-language readers tracking US-China AI competition, the diplomatic implication matters. Washington’s stated posture toward Beijing on frontier models is a race: out-build, out-deploy, out-regulate before the other side locks in a standard. In that frame, a domestic lab that insists on guardrails against autonomous-weapon use or domestic surveillance is a liability, because Beijing has no such constraint. The DoD designation was not about national-security risk in the traditional sense; it was about a single company saying no to a use-case the Pentagon wanted. The dinner is Trump’s way of saying: I will sit with you privately, but the answer to the “all legal purposes” question is already written.

Who Wins, Who Loses, What Happens on the 29th

Anthropic is preparing to file for a public listing. In a government where the president calls his favorites by name and tweets about export orders, a personal relationship with the occupant of the Oval Office is not a luxury; it is a risk-management strategy. Amodei gets a channel. The company gets, at minimum, a standing invitation to brief before the next enforcement action.

Trump gets a story. “I forgave the one company that told me no” is a better narrative than “my DoD froze a startup over a weapons clause” for the base that expects him to punish perceived disloyalty and reward reconciliation. It also lets him claim credit for pulling the most capable US lab back into the national AI sprint without publicly dismantling the safety constraints that distinguish Anthropic from its competitors.

The 29th brings the broader gathering: Trump and Speaker Mike Johnson sitting with the full CEO cohort. If Amodei is in the room, the message is that the DoD designation was a blip, already buried under a handshake. If he is not, the private dinner becomes a consolation prize, and the public meeting becomes a reminder that the administration’s tolerance for dissent in AI safety policy has a width of one.

Either way, the empty chair at Xi’s banquet did its work. It drew a line through the week’s calendar, and the line happened to run through the most contested question in American tech policy right now: who gets to decide what the machines are allowed to do, and which president’s handshake carries more weight than a federal agency’s memo.