Anthropic just modeled three versions of the American economy in 2030, and in the darkest one, unemployment tops 11.9 percent even as GDP grows by nearly a third.
On September 9, 2026, Anthropic's economics team published Scenarios for our Economic Future, a report that puts real numbers on a question the company has mostly dodged: what does its own technology do to the job market? The answer, in the most extreme of three scenarios, is grim. Cognitive unemployment hits 17.9%. Overall unemployment reaches 11.9%, worse than anything the US saw after World War II. GDP surges 32.4% to 44.4 trillion dollars anyway.
That's the tension running through the whole report. Anthropic doesn't try to hide it. The model rests on a companion technical paper by economists Anton Korinek, Charlie Jones, Petra Sacher, Alex Cotter and John McCrory. It lets anyone tune their own assumptions about how fast AI capabilities improve and how quickly companies adopt them. Set the dials low, and you get something close to the internet's impact on the economy: steady, unremarkable growth. Set them high, and you get a labor market that breaks.
In the modest scenario, GDP grows 1.6% to 34.1 trillion dollars by 2030, roughly in line with historical productivity trends. In the substantial scenario, AI handles about half of knowledge work autonomously, GDP jumps 8.3% to 36.3 trillion dollars, and knowledge worker wages stagnate while other workers see gains. The extreme scenario is where things get ugly: recursive, self-improving AI takes over most knowledge work without creating enough new tasks to replace it, and displaced professionals face prolonged unemployment.
Money keeps flowing. Just not to workers. Capital's share of income climbs from 40% today to 43.9% in the substantial scenario and 54.8% in the extreme one, meaning labor's share falls from roughly 60% to 45%. Knowledge worker wages drop more than 10% in that scenario. The economy ends up a third bigger. Workers, collectively, take home about what they would have earned without any of it.
Anthropic Researcher Jacob Coxon Quits Warning AI Could Kill Us All Jacob Coxon, a 27-year-old pretraining researcher, resigned from Anthropic on September 8, 2026, warning that AI labs are racing toward self-improving superintelligence they can't control. Anthropic's own Alignment Science Lead, Evan Hubinger, publicly agreed there's more than a 10% chance AI kills all humans within a decade. The warning lands as... - Anthropic researcher quits over AI safety concerns - AI superintelligence risk researcher leaves tech industry
The report tells displaced professionals to look at construction and nursing, fields it flags as far less exposed to automation. That's a real recommendation, not a hedge, and it says a lot about which jobs Anthropic thinks its own models are coming for first: analysts, writers, coders, anyone whose output is mostly language and logic on a screen.
The same week, a researcher quit warning about extinction #
Here's the thing that makes this report hard to read at face value. The same day it published, Bloomberg and NPR reported that Jacob Coxon, who led pretraining efforts at both OpenAI and Anthropic, resigned and told colleagues the industry is racing toward self-improving superintelligence it can't control. Anthropic's own internal safety teams reportedly estimate the odds of an AI-caused catastrophe at greater than 10% within the next decade, and the company still lacks a working plan to solve the alignment problem before more capable models arrive. David Sacks, meanwhile, has called for Anthropic to its reported 2 trillion dollar IPO over the same concerns.
So the company publishing warnings that its own technology might be uncontrollable is, in the same week, publishing detailed modeling of how that same technology reshapes the labor market by 2030. Frankly, you don't get to have it both ways forever. Either the technology is dangerous enough to threaten human control, in which case a GDP forecast is a strange thing to lead with, or it isn't, in which case the extinction warnings deserve more scrutiny than they're getting.
Anthropic's numbers also sit apart from the rest of the industry's forecasts. McKinsey has projected AI could add 13 trillion dollars to the global economy by 2030. Goldman Sachs is far more conservative, putting the boost from widespread adoption at roughly 7 trillion dollars in annual global GDP over a decade, with some of its own economists modeling closer to 1 trillion dollars a year. Anthropic's extreme scenario alone implies roughly 10.8 trillion dollars in added US GDP by 2030. No other major forecast pairs a number that large with an unemployment rate that would qualify as a depression.
None of the three scenarios is a prediction, and Anthropic says as much. The whole point of the interactive model is to let you set your own assumptions about AI's pace and adoption. Then you see what falls out. But a company doesn't build a tool that outputs 11.9% unemployment unless it thinks that outcome is at least plausible. That's not nothing. And it doesn't let a senior researcher walk out the door invoking extinction risk without that shaping how seriously readers take its growth numbers.
What Anthropic hasn't published is a plan for the millions of knowledge workers in its own extreme scenario. The report names construction and nursing as safer harbors. It doesn't say who retrains them, who pays for it, or how fast that has to happen if the substantial scenario, not the extreme one, turns out to be the floor.
Also read: OpenAI Puts AI Doomer Paul Christiano on Its Safety Board • How Context Window Length Actually Breaks Your AI Agent's Unit Economics • nCino Proves Banks Will Actually Pay for AI When It Works
A Lawsuit Says Anthropic's Claude Max 20x Delivers Far Less Than Advertised A federal class-action lawsuit accuses Anthropic of misrepresenting how much usage its $100 Max 5x and $200 Max 20x Claude plans actually deliver. Court filings show Max 20x provides only about 1.7 times the weekly Sonnet hours of Max 5x, despite costing double, and a hearing on Anthropic's motion to dismiss is set for November 6, 2026. - Claude Max 20x usage limits lawsuit against Anthropic - does Claude Max 20x deliver promised token capacity
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.