There is something unusual about an AI company releasing a public tool that tracks how its own technology might eliminate jobs.

That is exactly what Anthropic did on March 5. The company behind Claude, one of the world’s most widely used AI assistants, published a research paper and an interactive index that measures which white-collar jobs face the highest risk of being automated by large language models.

The tool is called the AI Exposure Index, and it does not pull any punches.

The Numbers

The research, authored by economists Maxim Massenkoff and Peter McCrory, introduces a metric called “observed exposure.” Instead of guessing which jobs AI might affect based on theory alone, the team combined three data sources: the O*NET database of roughly 800 US occupations, real-world usage data from Claude itself, and earlier research measuring whether an AI can make a given task at least twice as fast.

The result is a ranked list of how much of each job can already be handled by AI.

The most exposed occupations:

  • Computer programmers: 75% of daily tasks flagged as automatable
  • Customer service representatives: 70.1% coverage
  • Data entry keyers: 67.1% coverage
  • Medical record specialists: high exposure (exact figure varies by subspecialty)

The least exposed? Cooks, motorcycle mechanics, lifeguards, bartenders, dishwashers, and dressing room attendants. Basically, any job where you physically do things with your hands in unpredictable environments.

The “Great Recession” Warning

Fortune ran the story under the headline that a “Great Recession for white-collar workers” is possible. That is a reference to 2007-2009, when the US unemployment rate doubled from 5% to 10%.

But here is the nuance that matters: it has not happened yet.

The researchers found “no systematic increase in unemployment for highly exposed workers since late 2022.” In plain English, people in AI-vulnerable jobs are not losing their positions at higher rates than anyone else.

So why is this news?

Because something subtler is happening. Hiring of workers aged 22-25 in the most exposed occupations has measurably slowed. Companies are not firing people who already have jobs. They are quietly hiring fewer new ones.

If you are a 35-year-old programmer with a decade of experience, your job is probably fine for now. If you are a 22-year-old fresh out of college applying for your first programming role, you are walking into a different market than the one that existed two years ago.

The Gap Between What AI Can Do and What It Actually Does

One of the most interesting findings is that actual AI adoption in the workplace remains far below what is technically possible. As the researchers put it, “actual coverage remains a fraction of what’s feasible.”

This matters for two reasons.

First, it means there is still time. AI can theoretically handle 75% of a programmer’s tasks, but most companies have not restructured around that capability yet. Adoption takes years, not months. There are procurement cycles, training periods, regulatory concerns, union negotiations, and plain old institutional inertia.

Second, it means the impact could accelerate quickly once adoption starts. Right now there is a gap between AI’s capability and its deployment. When that gap narrows, and it will, the labor market effects could arrive faster than anyone expects.

Who Is Actually at Risk?

The demographics of AI exposure challenge a lot of assumptions. For decades, automation anxiety focused on factory workers, truck drivers, and retail employees. Blue-collar jobs. The kind of work that politicians campaign on protecting.

This time is different. The workers most exposed to AI displacement are:

  • Older (more experienced workers have more of the knowledge-work tasks AI can replicate)
  • Female (women are overrepresented in administrative, customer service, and healthcare documentation roles)
  • Highly educated (a graduate degree does not protect you; it actually correlates with higher exposure)
  • Better paid (the jobs AI threatens most are white-collar, professional-salary positions)

This is the opposite of every previous wave of automation. It is not the lowest-paid workers at highest risk. It is the ones with degrees and office jobs.

What AI Cannot Replace (Yet)

The zero-exposure jobs on the list tell their own story. Lifeguards, bartenders, cooks, mechanics. These are roles that require physical presence, unpredictable problem-solving, and real-time human interaction that AI simply cannot replicate.

If your job involves sitting at a computer processing information, analyzing text, writing reports, or answering questions, some portion of it can already be done by an AI model. If your job involves standing up, moving around, and dealing with physical objects and real humans in unpredictable situations, you are in better shape.

That does not mean those jobs pay more or are more prestigious. But it does mean the traditional assumption that “more education equals more job security” is breaking down.

Why Anthropic Built This

The obvious question is: why would an AI company build a tool that highlights the damage its own product might cause?

The cynical answer is that it is a PR move. Building a job-loss tracker makes Anthropic look responsible and transparent, which is good for business when regulators are watching.

The more charitable interpretation is that Anthropic genuinely believes the problem is real and that better measurement leads to better policy. As CEO Dario Amodei has said, he thinks artificial general intelligence could arrive “within one to two years.” If that timeline is even roughly correct, the window for preparing the workforce is very short.

The researchers framed it this way: they want to create “an established approach” that helps future observers “separate signal from noise” when monitoring real-time labor market changes. In other words, they want people to have good data before the situation becomes urgent, not after.

What You Can Actually Do

If you are reading this and wondering about your own job, here are some practical takeaways.

Check your exposure. Look at the list. If your daily work involves tasks that AI already handles well (writing, data analysis, customer communication, coding, documentation), you have meaningful exposure. That does not mean you will lose your job tomorrow, but it does mean your role will change.

Learn to work with AI, not against it. Workers who develop AI skills command significantly higher wages than peers who do not. The data consistently shows that the people who thrive in AI-disrupted fields are the ones who learn to use it as a tool rather than compete with it directly.

Watch the entry-level market. The clearest early signal in this data is the slowdown in junior hiring. If you are early in your career or about to enter the workforce, this is the most important trend to understand. Companies are not laying people off. They are simply not replacing them at the same rate. The impact shows up as fewer open positions, not as pink slips.

Do not panic, but do not ignore it. The gap between AI’s capability and its deployment gives everyone some runway. But that runway is getting shorter. The best time to adapt is before the trend becomes obvious to everyone.

The Bigger Picture

There is something quietly remarkable about this moment. An AI company has published a research tool that essentially says: “Here is exactly how our technology will reshape the labor market, and here is the data to prove it.”

Whether you think that is admirable transparency or calculated self-interest, the data itself does not care about motivation. The AI Exposure Index exists. The numbers are public. And for the first time, anyone can look up their occupation and see, with real data rather than speculation, how much of their work an AI can already do.

The question is not whether AI will change the job market. That is already settled. The question is how fast, and whether we use the warning signs in time.

The signal is in the data. The entry-level hiring slowdown is not a prediction. It is already happening.