Who Decides If AI Is Too Dangerous? (Spoiler: Nobody Really Knows)

Here's where it gets messy: Anthropic is one of the most serious AI companies out there—the kind that gives nice talks about safety. Then it releases Fable 5 and suddenly the American government (or at least parts of it) has opinions. The issue isn't that the model is inherently evil. The issue is that there's no universal mechanism to decide what counts as "too risky." It's like asking five different people to define when a car is too fast: one says 50 mph, another says 200, one says "it depends," and one hasn't answered yet.

In the Decoder podcast, Hayden Field—senior AI reporter at The Verge—digs into the question and finds something interesting: the tech community is fractured. Some think companies should self-regulate (spoiler: it never works). Others want government rules (spoiler: governments don't understand tech). Some argue it should all be open-source so "anyone can check" (spoiler: anyone includes bad actors too). Nobody is completely wrong, but nobody is completely right either.

The real question is: who gets to decide? The companies building it? Nope, they're biased. Governments? They're still figuring out what a hashtag is. Independent experts? Which experts, and who pays them? It's like trying to regulate the internet when half the people involved don't know how to turn on a computer.

Then Trump shows up. His administration enters the conversation and suddenly the tone shifts. It's no longer "Hey, let's write some ethical guidelines," it becomes "Who's in control here?" Politics, national security, global tech competition—it all blends together, and the discussion about what's "too dangerous" stops being about safety and starts being about power.

Field takes on the paradox: if you approve everything, you risk invisible monsters. If you ban everything, you kill innovation. If you say nothing, the problem doesn't disappear, it just goes underground. Meanwhile, companies keep building, governments keep holding meetings, and regular people keep using chatbots with zero clue what's happening backstage.

Field's conclusion—implicit but clear—is that the current system is broken. Not because the people involved are incompetent (though some are), but because the problem is structurally unsolvable as long as everyone plays by their own rules. We need a more honest conversation, less posturing. And we need it soon.