Two months ago, the White House called Anthropic a “radical left, woke company.” Last Friday, Treasury Secretary Scott Bessent and White House Chief of Staff Susie Wiles sat down with the AI firm’s CEO Dario Amodei for what the administration described as a “productive and constructive” meeting. According to Axios reporting, the two sides discussed “opportunities for collaboration” and the balance between innovation and safety.
The shift is jarring. It’s also entirely predictable.
What we’re watching is the collision between ideology and necessity. The Trump administration spent months feuding with Anthropic, with the president himself calling the company run by “left wing nut jobs” who were trying to “strong arm” defence. He declared the government would have no further business with them. Yet here they are, talking.
The reason is simple: Anthropic’s technology is apparently too useful to ignore.
The Leverage of Capability
Anthropic released Claude Mythos, a preview of what the company claims is an AI capable of outperforming humans at certain hacking and cybersecurity tasks. The tool can identify bugs in decades-old code and autonomously find ways to exploit them, according to Anthropic. A few dozen companies have been given access so far, and researchers have described it as “strikingly capable at computer security tasks.”
That’s the kind of capability governments want locked down. Or better yet, locked in.
The timing matters. Last week, Amodei said the company had “spoken to officials across the US government” and offered to work with them. The Friday meeting feels like the government’s response. Not a reversal, exactly. More like a recognition that whatever grievance exists with Anthropic’s politics or values, the firm’s technology is critical enough to warrant a seat at the table.
The Supply Chain Designation Nobody Wanted
In March, things came to a head. The Defence Department labeled Anthropic a “supply chain risk,” a designation that essentially means the government considers the technology insecure for official use. It was the first time a US company had been publicly given that label, which made it extraordinary.
Anthropic fought back in court, arguing the label was retaliation. Defence Secretary Pete Hegseth, Anthropic claimed, was punishing the company because Amodei refused to give the Pentagon unfettered access to its tools. The firm worried about mass surveillance and autonomous weapons development. A federal court in California largely sided with Anthropic’s reasoning, though a federal appeals court denied their request to temporarily block the designation.
But here’s the thing: according to court records, Anthropic’s tools are still in use at many government agencies. The supply chain risk label didn’t actually stop adoption. It just created friction and awkwardness.
When Politics Meets Pragmatism
That Friday meeting represents something deeper than a policy shift. It’s an admission that business and national security interests can override public antagonism. When Trump was asked about the meeting as he arrived in Phoenix, Arizona, he said he had “no idea” about it. Whether that’s literally true or diplomatic deniability doesn’t much matter. The meeting happened. The White House confirmed it. And the tone was cooperative.
What’s interesting is how little either side is claiming victory. Anthropic’s representative didn’t comment. The White House described the conversation in measured terms about “shared approaches” and “protocols.” Nobody is spiking the ball. Nobody is pretending this resolves the underlying tension between the administration’s ideology and its practical need for advanced AI capabilities.
The real question isn’t whether this meeting signals a permanent thaw. It probably doesn’t. Rather, it reveals something about how power actually works in practice: sometimes you need the people you disagree with, and sometimes you have to sit down with them anyway.
