Anthropic’s Amodei Refuses to Walk Back His AI Job Warnings

Anthropic CEO Dario Amodei used a new Bloomberg appearance to refuse the walk-back that has swept the rest of the AI industry, telling host Emily Chang on The Circuit that his warnings about AI-driven job loss have not changed and that critics accusing him of hyping the technology for profit are themselves running a marketing campaign. The interview, posted by Bloomberg Originals this month, places Amodei at the sharp end of a public split inside the labs building the systems, even as OpenAI’s Sam Altman and Nvidia’s Jensen Huang have both stepped away from their own dire forecasts.

Amodei has been the loudest AI executive on the employment question since he told Axios in May 2025 that AI could erase half of all entry-level white-collar jobs and push unemployment to between 10% and 20% within one to five years. The Bloomberg interview is the clearest signal yet that he is not joining the backpedal, and that he blames the social-media environment, not the data, for the way his message has been flattened into a slogan.

Anthropic’s Amodei Refuses to Walk Back His Job Warnings

Bloomberg Originals framed the episode around the quote it used as its lead: The idea that this is cheap marketing is itself cheap marketing. The line is Amodei’s rejoinder to a stream of accusations, traced through Jeff Jarvis and other commentators, that he runs a doom-marketing playbook to scare regulators into favoring his company. In its post announcing the interview, Bloomberg Originals wrote that Amodei “pushes back against critics who accuse him of hyping artificial intelligence risks to benefit his company,” and added that he “blasts Silicon Valley’s social media ‘disease’ and explains why society needs to start planning for the future of employment right now.”

The position inside Anthropic is that the original warning was never about promotion. The company has been laying out the threat publicly for more than a year, including a near-20,000-word essay from Amodei titled The Adolescence of Technology and a separate paper from the company arguing that AI can be more dangerous than the displacement risk alone, as Times of India reported. That is the same line of argument that critics such as David Sacks have hit from the other side, calling the regulatory framing a form of capture (the regulatory-capture case against Anthropic’s warnings).

Amodei’s response to that charge is procedural rather than predictive. In the Bloomberg appearance he points out that nearly every interview he gives, alongside the warning, also walks through possible policy responses. The list, as reported, runs to tax policy, macroeconomic measures, and the creation of new kinds of work. The argument is that society should be planning now, not after the wave hits, and that a CEO naming the risk is not the same as a CEO selling the risk (The Circuit episode on Amodei’s pushback).

The ‘Hump’ Argument for Why Productivity Is Not Safety

The core of the Bloomberg interview is a framework Amodei has been building toward for a year. In his framing, when AI automates 90% of a job, the remaining 10% is not a safe harbor. The worker still in the seat looks more productive at first, because the model is doing the bulk of the task, and that productivity gain is what people observe and call success. Amodei calls that the usual hump. The risk, as he laid it out to Axios, is that the curve does not flatten at 10%. Automation can keep moving toward 100%, and the people attached to the 10% that just disappeared are the ones the labor market will have to absorb (Amodei’s May 2025 warning on entry-level white-collar jobs).

The contrast he keeps drawing is between tasks and jobs. AI can replace tasks inside a profession long before it replaces the profession, and the gap between the two is where the policy choices sit. The earlier, more chilling version of the same idea was the line Amodei gave Axios: Cancer is cured, the economy grows at 10% a year, the budget is balanced – and 20% of people don’t have jobs. The 90/10 version is the gentler packaging of the same prediction, and it is the framing the Bloomberg interview is built on.

The idea that this is cheap marketing is itself cheap marketing.

That line lands in the Bloomberg interview at the moment Amodei is asked about the marketing critique, and the next beat is his complaint that social media has compressed a long argument into a clip. He did not name the company, the product, or a single policy in the lead-up, the way the critics have framed him. He named the practice.

OpenAI, Nvidia, and the New Reassurance Chorus

The reason the Bloomberg appearance matters is that almost everyone else in the lab business has moved. Sam Altman, whose earlier language included AI probably replacing most of the jobs people do today and entire categories being totally gone, said at a Commonwealth Bank of Australia conference in Sydney on May 26 that the white-collar apocalypse is not coming, and that he is delighted to be wrong about it. He told the Sydney audience he had expected more impact on entry-level white-collar jobs by now than has materialized, and that his intuitions on the question were just off (Altman’s ‘delighted to be wrong’ reversal in Sydney).

Jensen Huang has gone further. Speaking to Channel News Asia in late May, the Nvidia chief executive called the AI-to-layoffs narrative too lazy to be worth the airtime, and asked aloud how CEOs could be claiming AI cost jobs two years ago when the technology only became broadly useful six months ago. His line, reported in France 24, was that executives were using the AI label to sound smart and that the industry was scaring people in a way he finds irresponsible (Huang calling the layoff-AI narrative too lazy).

That puts Amodei and Huang on opposite sides of a public argument Huang first opened a year ago, when he said he pretty much disagrees with almost everything Amodei says. The deeper split is structural. Anthropic is heading toward an IPO that the Financial Times reports is targeting a $900 billion valuation on a fresh $30 billion round, and OpenAI is aiming for $280 billion in revenue by 2030, up from $25 billion today. The investment case rewards a softer employment message, which is one reason France 24’s piece explicitly tied the walk-backs to the upcoming IPO calendar. Federal Reserve Governor Lisa Cook, speaking at Stanford in late May, framed the same moment in starker institutional language: We could be approaching the most significant reorganisation of work in generations, and AI-related job losses could precede any gains.

Amodei is betting that the public and the policymakers will read the labor data more carefully than the executives are. The next two quarters of hiring reports will be the test.

The Layoff Math Has Not Cooperated

Even with the reassuring turn at the top of the industry, the actual layoff numbers through the first five months of 2026 do not match the softer message. Outplacement firm Challenger, Gray & Christmas tracked AI cited as a factor in nearly 50,000 job cuts through April, and the firm’s Andy Challenger, workplace expert and chief revenue officer, summed up the dynamic in a sentence the body of the industry keeps proving out: technology companies are leading all industries in layoff announcements and are often citing AI spend and innovation, and the money for those roles is going away, whether the job itself is being replaced or not. The visible cuts so far in 2026 include Meta’s roughly 8,000 roles eliminated in May, about 10% of its workforce, and Intuit’s 17% staff cut, or 3,000 people, the same month. Amazon, Alphabet, Microsoft, Walmart, and CrowdStrike have all named AI as a factor in round after round through the spring.

• AI cited as a factor in nearly 50,000 job cuts through April 2026 (Challenger, Gray & Christmas).
• Meta eliminated roughly 8,000 roles in May 2026, about 10% of its workforce.
• Intuit cut 17% of its staff, or 3,000 people, in May 2026.
• Stanford/Brynjolfsson ADP payroll research: a 13% relative decline in employment for workers aged 22 to 25 in the most AI-exposed jobs since generative AI took hold, with older workers in the same occupations flat or growing.
• MIT NANDA initiative: of 300 enterprise AI deployments studied, only about 5% delivered rapid revenue gains, and the rest produced little or no measurable impact on the bottom line.

The Stanford read is the one most often cited as proof that the apocalypse is not landing. The damage is showing up at the entrance, not the exit: AI is good at the codified, book-learning tasks a new graduate is hired to do, and bad at the tacit judgment that experience buys, so companies are quietly not opening the junior roles beneath their existing staff. That is exactly the pattern Amodei warned about in 2025, and the one the Bloomberg interview is now extending (what the Stanford and MIT data actually show).

From Warning to Policy

The second pillar of Amodei’s argument is the policy one. In the Bloomberg interview, and in the longer Anthropic essays that followed, he points out that he pairs the warning with a list of responses he says he discusses in nearly every appearance. The three that surface most often are tax policy on the gains from automation, macroeconomic measures to absorb the transition, and the creation of new kinds of work. The argument is that a CEO naming a risk while also naming the responses is not the same as a CEO issuing a prophecy and walking away, and that the doom-marketing charge is an attack on a policy paper, not a forecast.

That is also why the IPO timing matters. Anthropic’s confidential S-1 was filed on the road to a valuation reported in the press at $965 billion, days after Claude Code hit $2.5 billion in run-rate revenue, and the company has been pushing for enterprise lock-in against OpenAI and Microsoft in the same window. The doom-marketing charge lands hardest on a company with the most to gain from a heavy regulatory environment, which is one reason the regulatory-capture critique and the enterprise-pricing fight are running in parallel (Anthropic’s $965B IPO and the enterprise pricing fight).

Outside the lab business, the productivity argument runs the other way. Cognizant CEO Ravi Kumar said at Davos he sees $4.5 trillion in U.S. productivity gains waiting to be captured even if AI capabilities stop advancing, and EY’s Raj Sharma told Fortune his firm has used AI agents to open a $6 billion tax-services market it could not profitably serve before. Both executives are betting the same way on the demand side that Amodei is betting against on the supply side, and the next twelve months of earnings calls will decide which framing the public buys.

Where the Line Sits Now

Three positions are now in play, and none of them is going away. Sam Altman has admitted his 2025 intuitions on white-collar displacement were off, and is now selling the line that the human part of work is durable. Jensen Huang has called the AI-to-layoff connection lazy and irresponsible, and is now selling the line that the productivity story is the real story. Dario Amodei has held the line that the labor market is in the early phase of a multi-year reorganization, and is selling the line that society needs to start planning the response before the phase ends. The IPO calendar the Financial Times flagged for Anthropic, at a valuation reported around $900 billion, and the public tour the next set of earnings will fund, will tell the market which of those three framings the data has vindicated by the end of 2026.

Frequently Asked Questions

What did Dario Amodei say in the latest Bloomberg interview?

On The Circuit with Emily Chang, Amodei said his concerns about AI-driven job loss have not changed, called the productivity gains workers are seeing today the usual hump, and accused critics who frame his warnings as cheap marketing of running their own cheap-marketing campaign. He blamed social media for collapsing a long policy argument into a clip, and pointed listeners to his longer essays for the policy responses he says he discusses in nearly every interview.

Why is he pushing back at critics now?

Because the consensus inside the AI industry has shifted against the kind of warning he has been issuing since his May 2025 Axios interview. Sam Altman publicly walked back his own forecasts in late May, Jensen Huang called the AI-to-layoff narrative irresponsible, and the regulatory-capture charge has resurfaced as Anthropic’s IPO has come into view. Amodei is using the Bloomberg slot to mark the distance between his position and the new reassurance chorus, and to keep the policy debate attached to the forecast.

Have other AI CEOs walked back their warnings?

Yes. Sam Altman told a Commonwealth Bank of Australia conference in Sydney on May 26 that the jobs apocalypse some of the companies in our space advocate or talk about is not coming, and that he had expected more entry-level white-collar impact by now. Jensen Huang, speaking to Channel News Asia in late May, dismissed executives who blamed AI for layoffs as too lazy to do the work, and said the industry is scaring people in a way he finds irresponsible. Both are now several steps away from the language they were using in 2025.

What does the job-cut data show so far in 2026?

Outplacement firm Challenger, Gray & Christmas tracked AI cited as a factor in nearly 50,000 job cuts through April 2026. Meta eliminated roughly 8,000 roles in May, about 10% of its workforce. Intuit cut 17% of its staff the same month. Stanford research using ADP payroll data found a 13% relative decline in employment for workers aged 22 to 25 in the most AI-exposed jobs, with older workers in the same occupations flat or growing.

Is the productivity bump real or just a temporary phase?

Real in the short term, contested in the long term. Uber, Walmart, and several enterprise software companies have described measurable productivity lifts from AI in the past year, and Cognizant says there is $4.5 trillion of U.S. productivity on the table even if the models stop improving. Amodei’s argument, which he lays out in the Bloomberg interview, is that the productivity bump is the hump that masks the next stage of automation, and that the policy response has to arrive before the bump ends.