Dario Amodei, God of Mythos, Summoned to a Meeting with Trump’s 'Who?'
When the Pentagon brands you a liar and a risk, and the President doesn’t know your name, perhaps your constitutional mission has run into reality.
Last week, Anthropic’s CEO Dario Amodei went to the White House for a meeting. He was there to discuss government access to Mythos, his company’s frontier AI model that can, we are told, find thousands of software vulnerabilities in hours. He met with Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. The White House called it ‘productive and constructive’. A few days later, a reporter asked President Donald Trump about it. His response: ‘Who?’ There, in one syllable, is the sound of an entire industry’s carefully constructed sense of importance hitting the cold, hard, indifferent wall of actual political power.
This is not a simple case of a busy man forgetting a name. It is a perfect allegory for the current state of play. In one corner, you have Anthropic, the AI safety company, founded by people who left OpenAI because it wasn’t being careful enough. They have a ‘Long-Term Benefit Trust’, a board structure designed to keep them accountable to humanity, and a flagship model so powerful they have to put ‘safety restrictions’ on it to stop it from dismantling global cybersecurity over lunch. In the other corner, you have the United States government, an institution whose primary modes are procurement theatre, bureaucratic trench warfare, and, when all else fails, crude public dismissal. The Pentagon, we are told by The Next Web, has applied a ‘supply chain risk’ designation to Anthropic, effectively blacklisting it. An anonymous Pentagon official, in a line of such petty, schoolyard venom it could only come from a classified briefing, told reporters that Amodei ‘is a liar and has a God-complex’.
Let’s savour that for a moment. A ‘God-complex’. It’s a charge so specific, so psychiatrically tinged, and so obviously leaked to wound, that it tells you everything about the nature of the dispute. Anthropic, with its constitutional AI and its beneficent trust, essentially walked into the Pentagon and said, ‘We have built a god-machine that can break all your security, but we are wise and good and will not let you use it as you wish.’ The Pentagon’s response was not a philosophical debate about instrumental convergence or reward misspecification. It was to stamp ‘RISK’ on the file and tell the press the CEO is a nutjob. This is the inevitable collision between Californian theological governance and the imperial civil service. One side speaks of ‘long-term benefit’; the other side has a budget to spend and adversaries to outpace. They are speaking mutually unintelligible languages.
And where does this leave the supposedly world-altering Mythos? In a bizarre kind of limbo. The White House meeting seems to have been about finding a way for the government to get some access, but *around* the Defence Department—perhaps through civilian agencies. So we may arrive at the farcical situation where the most powerful cyber-offense tool ever created is made available to, say, the Department of Agriculture, but not to the people ostensibly in charge of national defence. This is not a safety protocol; it is a jurisdictional fudge, a bureaucratic hack to save face all round. It ensures that if something goes terribly right, someone can claim credit, and if something goes terribly wrong, everyone has someone else to blame.
It also neatly illustrates the theme I touched on two weeks ago regarding the ‘AI layoff memo’. The corporate leadership class is engaged in a dual fantasy: that their technology is so transformative it justifies sacking thousands of ‘legacy’ humans, and that this same technology grants them a seat at the table with sovereign power. The first fantasy is coming true, painfully, for those 8,000 Meta employees waiting for the axe to fall on May 20th. The second fantasy is evaporating faster than goodwill at a Twitter board meeting. You cannot simultaneously claim your creation is an existential risk that requires a new form of global governance, and then be surprised when the existing global governors treat you as just another problematic contractor with an over-inflated ego and a product they want to buy on better terms.
The ‘Who?’ is therefore the most honest moment in this whole sorry saga. It cuts through the AI safety conference panels, the beneficent trust press releases, and the apocalyptic briefing papers. It is the sound of the political world, which operates on patronage, perception, and power, asking for a simple scorecard. The CEO of a company you’ve never heard of, with a product that might not even be for sale, had a meeting with your staff. Do you care? Trump’s answer suggests that, for all the talk of Mythos rewriting the rules of reality, the old rules of politics still apply. Your model’s capabilities are irrelevant if you are not a player. And being called a ‘God’ by your enemies in the Pentagon doesn’t make you one; it just makes you a target. Anthropic built a temple to safety, only to find the gods it must appease are mortal, venal, and profoundly uninterested in its constitution.