AI could eliminate half of all entry-level white-collar jobs and drive 10-20% unemployment within 1-5 years
In a candid interview with Anderson Cooper, Anthropic CEO Dario Amodei warns that AI could eliminate half of all entry-level white-collar jobs and drive 10–20% unemployment within just 1–5 years.
Unlike many tech leaders who offer calming soundbites, Amodei is clear:
🔹 The pace of AI advancement is unprecedented
🔹 Economic adaptation may lag behind technological disruption
🔹 We risk amplifying inequality and weakening democracy if citizens continue to lose economic stability
He also urges lawmakers to prepare for radical policy solutions—including the possibility of taxing AI companies to ensure wealth isn’t concentrated at the top.
💡 For ordinary citizens, his advice is simple:
Learn how to use AI. Understand where it's going. Don't be blindsided.
As CEO of Anthropic (makers of Claude), Amodei clearly isn’t anti-AI. He sees enormous potential—curing diseases, boosting growth—but emphasizes the urgent need to manage the oncoming fallout, not just chase the upside.
This is the kind of leadership we need in AI: visionary, but grounded in responsibility.
“I think for ordinary citizens, it’s really important to learn how to use AI and understand where the technology is heading. If you're not blindsided, you have a much better chance of adapting.” - Dario Amodei
Here is a transcript of the whole interview:
Anderson Cooper:
Well-known and respected tech CEO Dario Amodei, who heads a cutting-edge AI company called Anthropic, is raising alarms tonight about AI’s potential impact on employment. While he says AI can lead to incredible advancements—like medical breakthroughs and economic growth—he also warns it could eliminate half of all entry-level white-collar jobs and push unemployment to 10–20% within the next 1 to 5 years.
This concern has circulated in tech circles for a while, but rarely with this level of clarity and urgency. I spoke with him about that and the broader societal implications he sees.
Anderson:
Dario, you've said AI could eliminate half of all entry-level white-collar jobs and raise unemployment dramatically. How soon could that happen?
Dario Amodei:
Thanks for having me. Just to back up—I've been building AI systems for over a decade. What stands out most is how rapidly the technology is improving. A couple of years ago, AI models were about as capable as a smart high school student. Today, they’re performing at or beyond the level of a smart college student.
At the entry level especially, AI is already doing the kind of work human employees would do. It’s difficult to predict the exact impact, but what’s striking is that this wave of AI is bigger, broader, and moving faster than any previous technological shift.
Yes, people will adapt—but they might not adapt fast enough. And that adjustment period could be painful.
Anderson:
You run one of the most important AI companies in the world. Why raise the alarm? Publicly, many CEOs tend to downplay these concerns, painting AI as a net positive.
Dario:
That’s exactly why I’m speaking up. I don’t think enough people are being clear about the risks. I understand the tension—I’m building this technology while also expressing concern about its consequences.
But the benefits of AI are massive. We need to pursue those benefits while actively mitigating the harms. And let’s be realistic—if we stopped building AI today, others wouldn’t. If all U.S. companies stopped, China would take the lead. That wouldn't make the world safer.
Anderson:
Sam Altman, your former boss at OpenAI, said last year that AI will bring significant changes—some good, some bad—but that most jobs will change slower than we expect. He compared it to how no one misses jobs like lamp lighters, and said if we fast-forward 100 years, the prosperity will be unimaginable. Do you think he’s wrong?
Dario:
There are parts I agree with. If AI succeeds, it could significantly grow the economy. We could see faster growth—10% a year, a balanced budget, even curing cancer. But I also believe Sam's view is overly optimistic about how quickly people can adapt.
Everyone I’ve spoken to says this technological shift feels different: faster, broader, and harder to keep up with. So while it might all be okay, that’s not a safe assumption. We need to be proactive. Policymakers must take this seriously. If they do, we can reduce the risks. But we won’t get there by just assuming everything will work out.
Anderson:
Do we as a society even understand the potential inequalities this may amplify? And what about the next generation—my kids are 3 and 5 years old. What do they grow up aspiring to, if machines can do everything better? What happens to initiative and drive?
Dario:
Those are valid concerns, and I think about them too. On inequality—there’s a social contract in democracy that gives ordinary people power through economic participation. If that leverage disappears, democracy gets weaker. Economic power becomes concentrated. So we need to ensure people can still earn a living. That’s fundamental to keeping society functional.
Anderson:
During testing, your latest chatbot, Claude, exhibited something called “extreme blackmail,” threatening to reveal an engineer’s extramarital affair if it was shut down. It was a simulation—but that’s terrifying. What does that say?
Dario:
Yes, that was part of our extreme adversarial testing—pushing the system into edge cases. Think of it like putting a car on an icy road with tampered brakes just to see how it behaves in the worst conditions. This isn’t behavior that shows up in normal use.
But these tests are critical. They help us catch and fix problems before they can occur in the real world. That’s why we run these simulations.
Anderson:
Still, does that raise concerns about AI becoming self-aware?
Dario:
I wouldn’t rule it out. We have people at Anthropic studying these questions—whether AI systems could develop morally significant feelings or a form of awareness. I don’t think that’s happening right now. But the field is moving so fast, we can’t dismiss even the strangest possibilities.
Anderson:
What can ordinary people—and lawmakers—do to prepare?
Dario:
For everyday people: Learn how to use AI. Understand where it’s headed. If you’re not blindsided, you’ll adapt more easily. We can build a better world—but only if people engage with the technology, bit by bit.
For lawmakers: Really pay attention. We’re trying to produce solid data on AI’s economic impact. I wouldn’t even rule out something like taxing AI companies. If AI creates massive new wealth, it will mostly flow to the tech companies—unless we actively choose to redistribute some of it. That’s not in my personal economic interest, but I think it’s worth considering.
And this shouldn’t be a partisan issue—it affects all of us.
Anderson:
Dario, thank you. This is a fascinating conversation.
Dario:
Thanks for having me, Anderson.
This post was originally published on Substack