Anthropic, the company that makes Claude, just released an interactive model showing that in its most extreme scenario, AI could throw nearly 14% of American workers out of jobs, with less than half of them finding new ones. They did not say this was unlikely. They said policymakers should "get ready to spend."
The Tool That Shows You How Bad It Could Get
The model is framed as a public service. You plug in your assumptions about how capable AI gets, how fast companies adopt it, whether it works alongside humans or simply replaces them, and then you watch what happens to GDP and employment. It is, in a way, a choose-your-own-apocalypse.
According to NPR, the tool spans a wide range of outcomes. At the gentle end, AI is a modest productivity boost, something like a really good spreadsheet program, and life goes on mostly as before. At the far end, GDP grows at more than seven times its current pace, which sounds incredible right up until you check what happens to workers at that same scenario. Nearly one in seven loses their job. Less than half of them find another one.
Anthropix's blog post puts it bluntly: in the extreme case, "AI transforms the economy" in ways that go "beyond any event in history, both in terms of magnitudes and in terms of the speed at which change happens." No historical analogy. Bigger than industrialization. Bigger than the internet. Bigger than anything they could point to as a reference. That is either a sales pitch or a warning. Possibly both.
Nobody Is Actually Saying Which One Happens
Here is the part where you might expect the experts at a major AI company to tell you which outcome they think is more likely. They did not do that. As NPR reports, the authors of the study explicitly decline to offer a prediction.
Anthropix co-founder Jack Clark told NPR he thinks the technology keeps improving fast but that spreading it through the actual economy will be harder than the optimists expect. "It will get really, really good," he said. "But it will make its way into the economy more slowly." Which is sort of reassuring, until you remember that the tool he helped build still has a scenario where seven percent of the country's workforce can't find new jobs after AI takes their old ones.
Anthropix's head of transformative AI economic studies, Anton Korinek, offered the kind of statement that sounds reasonable and terrifying simultaneously: "If the AI can do amazing things but nobody uses it, then it's not going to have an economic impact." True. Also: companies are very motivated to use it.
What 11,000 Regular People Think Is Coming
Anthropic didn't just model the economics. They surveyed nearly 11,000 people about their expectations, and according to NPR, the public lands somewhere in the messy middle. Average respondents expect a meaningful boost to productivity and growth. They also expect significant disruption to workers in fields that AI is likely to hit hardest.
Which means regular people, without economics degrees or access to proprietary AI benchmarks, have basically triangulated to the same uncomfortable conclusion the model produces: this is probably going to help the economy, and it is probably going to hurt a lot of workers, and those two things will be happening at the same time, in ways that do not automatically cancel each other out.
That is not a comfortable place to sit. It is, however, an honest one.
The 'Policymakers Should Get Ready to Spend' Part
Clark's comment to NPR about government response is worth sitting with for a moment. In the extreme GDP growth scenario, he said, policymakers would have "moves available" to support displaced workers that are "unimaginable today." His advice: "Policymakers should get ready to spend."
Set aside for a second that the current U.S. federal government has spent the better part of the last two years trying to cut, not grow, its capacity to support workers through disruption. Set aside the gutted labor agencies, the reduced social safety net, the general posture of the administration toward anything resembling a robust government response to economic displacement. Just take the advice on its face.
The premise is: AI creates a disruption so massive it has no historical parallel, millions of people lose jobs faster than they can find new ones, and the government's job is to catch them with spending. That is a plan that requires a government that wants to catch people. Whether that government exists is a separate and very important question.
The Dingo Take
Let's name what this tool actually is. It is an AI company publishing a model that shows, in graphic and interactive detail, that its own product could contribute to the worst unemployment shock in recorded history, and then declining to say that won't happen, and then asking policymakers to be ready to write very large checks. That is a remarkable thing to put your name on. It is also, in a strange way, more honest than most of what comes out of the tech sector, which typically promises disruption while marketing it as pure uplift.
The problem is that releasing an interactive tool and writing a blog post is not the same as having a plan. Anthropic can model seven different economic futures, survey eleven thousand people, and put Jack Clark on the phone with NPR, and none of that does a single thing to build the policy infrastructure that would actually protect workers if the bad scenario lands. That work happens in legislatures and agencies and budget negotiations, most of which are currently occupied with other priorities, or actively hostile to the concept of a government that protects workers from market disruption.
So yes, Anthropic deserves some credit for not pretending this is consequence-free. But credit for honesty is not the same as credit for responsibility. The company is building the technology in either scenario. They will profit from it in either scenario. Releasing a tool that shows how bad it might get is not a substitute for the much harder work of making sure the bad scenario doesn't eat the people who can least afford to be eaten.
Comments