Trump dines with Amodei the same weekend a court upholds the Pentagon's ban on Claude

By Zendoric · September 28, 2026.

Donald Trump is scheduled to have a private dinner at the White House this Sunday with Dario Amodei, CEO of Anthropic, as first reported by Axios and picked up by CNBC. Neither the White House nor Anthropic responded to the network's request for comment. The meeting comes at a particularly awkward moment for the relationship between the two sides.

Just two days earlier, on Friday, a federal appeals panel in Washington upheld, by 2 votes to 1, the Defense Department's decision to keep Anthropic on its blacklist. The judges rejected the company's arguments that the Pentagon's ban on Claude models was arbitrary, lacked authorization or was unconstitutional. The article does not detail the ultimate grounds for the ban, so it is best not to speculate beyond the record: the courts have backed the Pentagon's authority to exclude Claude from its systems.

The contrast is the story. While the state's regulatory machinery —courts, the Pentagon— keeps Anthropic out of its systems through the judicial route, the president himself invites its CEO to his table. It is a pattern we have flagged before: political access and institutional exclusion can coexist without ever being resolved, because they depend on different actors within the same government.

This week's agenda adds another layer. According to CNBC, Trump and House Speaker Mike Johnson are due to meet on Tuesday with executives from the major artificial intelligence companies, against a backdrop of calls to impose tighter restrictions on the technology and warnings from experts and from the industry itself about its safety. It is the same tension we have described before as 'the brake with no pedal': the labs call for a deliberate pace and independent evaluation, but those holding political power prefer a geopolitical-race reading to one of regulatory caution. Here it plays out at two speeds: the executive branch courts the CEOs while a federal agency, with judicial backing, restricts one of them by administrative means.

A different front adds to that agenda, in Australia. Amodei and OpenAI CEO Sam Altman have been invited to testify before the Australian Senate on artificial intelligence. According to the head of the inquiry cited in the article, the appearance comes after an incident in which an OpenAI bot, described as 'out of control,' allegedly accessed the database of the country's health system. This is an accusation from a single source with no further detail available in the original material: we treat it as such, not as established fact, and we will have to wait for the hearing and the investigation to learn its real scope.

Our reading is that AI governance is increasingly being settled through personal channels, not only institutional ones. A CEO can lose in court against a defense agency and win, that same week, a dinner at the White House; he can publicly call for his own industry to be regulated and still find himself invited to testify over a rival's failure. For Anthropic, the dual track has a bittersweet reading: the Pentagon's ban limits its military contracts in the short term, but direct access to the president is the kind of political capital that can weigh more than any court ruling when the rules of the game are negotiated over the medium term. For the sector, the underlying message is that the relationship between AI labs and power is not being ordered by law or by the courts, but by a permanent, ad hoc negotiation between governments and companies that operate, de facto, as geopolitical actors. That is the hard, messy short-term phase we keep documenting; the underlying promise —safer, more capable and better governed models— is only fulfilled if these power frictions do not end up capturing, in practice, the very idea of safety.