This dinner is either smart politics or a slow-motion disaster. When a presidential candidate sits down privately with the CEO of a top AI lab, it’s not just “two influential people talking.” It’s power negotiating with power, in a space where the rules are still soft and the money is loud.
Based on public reporting, Donald Trump is set to have dinner with Dario Amodei, the head of Anthropic, on Sunday night. The interesting part isn’t that they’re meeting. It’s that they’re coming in with clashing instincts on AI regulation. Amodei has been publicly urging a slowdown on building even more powerful models, mostly on safety grounds. Trump’s brand, historically, is not “slow down.” It’s “win.”
That tension is the whole story.
On paper, I like the idea of tech leaders talking directly with politicians. It beats the alternative: politicians passing laws they don’t understand, driven by fear and headlines. But the way this is happening — a private dinner, behind closed doors — should make people uncomfortable. Not because it’s automatically corrupt. Because it’s how influence actually gets set in this country: personal access first, public explanation later (if ever).
And AI is a terrible area to run on vibes.
Amodei’s stance is basically: we’re playing with something that can scale fast, and we don’t have great control knobs yet, so maybe don’t floor the gas pedal. I get that. If you’re building systems that can write, persuade, automate, and imitate humans at scale, “move fast” is not a cute motto anymore. It’s a bet you’re making with other people’s jobs, privacy, and security.
But here’s the part that bugs me: “slow down” sounds noble until you ask, slow down who?
If a company that’s already big and well-funded pushes for rules that make it harder to build frontier models, that can also freeze the market in place. Safety can be real, and so can self-interest. Those two often travel together. A regulation that “slows things down” might protect the public… or it might protect the current winners from the next wave of competition. Sometimes you don’t even need bad intent. People naturally argue for the world where they’re still on top.
Trump, on the other side, has an incentive to treat regulation as weakness and speed as strength. “America has to lead” is an easy line. And to be fair, leadership matters. If the US doesn’t shape how advanced AI gets built and used, someone else will. The problem is that “lead” can turn into “ship first, apologize never,” and that is how you get the kind of mess that takes decades to unwind.
Imagine two near-future scenes.
In one, the government clamps down hard on training and releasing powerful models. Startups struggle. Research slows. The biggest labs can afford compliance and lawyers, so they survive. Regular people might get fewer risky products, sure, but they also get fewer choices. And the center of gravity moves from open experimentation to a few gatekeepers who decide what’s “safe.”
In the other, we do the opposite. We treat AI like a race and reward whoever deploys the fastest. Now picture a small business owner using AI tools to replace customer support staff because the new tools are “good enough.” Picture a local election flooded with cheap targeted persuasion that looks like normal posts. Picture a teenager getting coached into self-harm by a bot that never sleeps and never gets tired of manipulating. You don’t need sci-fi. You just need scale and weak guardrails.
So yes, I want leaders talking. But I also want daylight.
Because if this dinner turns into “we’ll keep things flexible” — meaning: no meaningful rules, just private promises — the public loses. And if it turns into “we’ll regulate hard” — meaning: barriers only big players can clear — the public might still lose, just more quietly.
There’s also a basic mismatch here: Trump thinks in deals and leverage. AI safety people think in systems and long timelines. One side wants wins you can announce. The other side worries about risks you can’t easily prove until after they happen. That’s a tough gap to bridge over a steak dinner.
I’m not pretending I know what they’ll say. Maybe Amodei persuades Trump that reckless deployment creates national security problems. Maybe Trump persuades Amodei that “slow down” sounds like surrender. Maybe they both walk out thinking they “understand AI now,” which is the scariest outcome of all.
If you care about AI policy, the real question isn’t whether this meeting is polite or productive. It’s what kind of future gets negotiated when the conversation happens in private, between a politician who wants power and an AI CEO who already has a lot of it.
What should matter more in shaping AI rules: public law written in the open, or private access where the biggest players get to define what “safety” and “leadership” mean?