AI Can Do the Work, But It Isn’t Replacing Workers Yet, Anthropic’s Labour Study Shows

AI and the Labour Market: Early Signals Without a Shock?

Artificial intelligence is moving quickly into offices, code editors, customer service desks and research workflows. Naturally, the question that keeps surfacing across industries is simple: Will AI take away jobs? A new research report by Anthropic AI titled “Labour Market Impacts of AI: A New Measure and Early Evidence,” published on March 5, 2026, attempts to answer that question using a new measurement framework. The study introduces a concept called “observed exposure,” which combines theoretical AI capability with actual real-world usage data to understand how AI is interacting with jobs today. The conclusion, at least for now, is nuanced. AI has not yet caused a broad increase in unemployment. But there are early signals of slower hiring in some professions, particularly among younger workers entering the labour market.

The findings come at a time when AI adoption is accelerating globally, including in India’s rapidly expanding digital economy.

Why Previous Predictions About Job Losses Were Often Wrong?

Forecasting technological disruption has never been straightforward. The study points out that previous attempts to predict labour market shocks have frequently overestimated the speed and scale of change.

For instance, earlier research on job offshoring predicted that nearly a quarter of American jobs were vulnerable. A decade later, many of those occupations continued to grow normally.

Similarly, debates around industrial robots and the China trade shock have produced conflicting results among economists even years after the events took place.

This historical uncertainty is why the researchers approach AI cautiously. Rather than predicting future job losses outright, the study proposes a framework to track how AI exposure evolves over time.

The goal is simple: measure the change early, before the disruption becomes obvious.

Introducing “Observed Exposure”: A New Way to Measure AI’s Impact?

At the heart of the research is a new metric called observed exposure.

Instead of asking what AI could theoretically do, the researchers ask a different question: Which tasks are actually being performed using AI in professional environments?

To answer this, the study combines three major datasets.

First, it uses O*NET, a large occupational database that lists tasks performed across roughly 800 professions in the United States.

Second, it incorporates real usage data from AI systems, specifically from the Anthropic Economic Index, which tracks how people are using AI tools in real work settings.

Third, it builds on earlier academic research that evaluates whether a task can theoretically be sped up by large language models.

In that earlier framework, tasks are given a score:

• 1. If AI alone can double the speed of a task
• 0.5 if AI requires additional tools or software
• 0 if AI cannot assist meaningfully

By combining these elements, the study identifies occupations where AI is both capable and actively being used.

Is AI Still Far From Its Full Potential?

One of the most striking findings is that AI usage today is far below its theoretical capability.

Even in professions where AI could theoretically perform many tasks, real adoption remains partial.

For example, in computer and mathematics occupations, AI tools could potentially assist with about 94 percent of tasks. Yet actual usage currently covers only about 33 percent of those tasks.

This gap exists for several reasons.

Some tasks require human verification, legal approval or integration with specialised software systems. Others are slowed by regulation, organisational hesitation or workflow complexity.

A good illustration comes from healthcare administration tasks. Authorising prescription refills could theoretically be assisted by AI, but real-world usage remains limited because of regulatory requirements.

In short, capability does not equal adoption.

Which Jobs Are Most Exposed to AI Right Now?

The study identifies several occupations where AI exposure is already significant.

Among the most exposed roles are:

• Computer programmers
• Customer service representatives
• Data entry operators
• Financial analysts

Computer programmers appear at the top of the exposure ranking, with around 75 percent of tasks already covered by AI usage in some form.

Customer service roles follow closely, especially as many companies increasingly deploy AI-powered chat systems and automated workflows.

Data entry jobs are also highly exposed because reading documents and transferring information into digital systems is a task that AI can perform efficiently.

However, exposure is not uniform across the workforce.

About 30 percent of occupations show zero measurable AI coverage, largely because their tasks are physical or location-dependent.

Examples include:

• cooks
• lifeguards
• bartenders
• motorcycle mechanics
• dishwashers
• dressing room attendants

These roles involve manual work that AI cannot replicate easily.

Will Jobs With Higher AI Exposure Grow More Slowly?

The research compares its exposure measure with employment projections from the U.S. Bureau of Labour Statistics, which forecasts occupational growth between 2024 and 2034.

The relationship between AI exposure and job growth appears modest but noticeable.

According to the study, for every 10 percent increase in AI exposure, projected job growth declines by about 0.6 percentage points.

The effect is not large, but it suggests that AI could gradually influence hiring patterns over the coming decade.

Interestingly, this correlation only appears when real usage data is included. Theoretical capability alone does not predict employment changes effectively.

Who Is Most Likely to Work in AI-Exposed Jobs?

The study also examines the characteristics of workers in the most exposed occupations.

Compared with workers in low-exposure jobs, those in high-exposure roles are:

• more likely to be female
• more highly educated
• higher paid on average
• more likely to hold graduate degrees

For example, 17.4 percent of workers in the most exposed occupations hold graduate degrees, compared with only 4.5 percent in the least exposed group.

Average earnings are also significantly higher, with exposed workers earning about 47 percent more on average.

This means AI exposure is concentrated primarily among white-collar professions, rather than traditional manual labour.

Has There Been No Major Increase in Unemployment So Far?

Despite rising AI adoption since the launch of generative AI tools in late 2022, the study finds no clear increase in unemployment among highly exposed workers.

Using data from the Current Population Survey, researchers tracked unemployment trends for workers in the most exposed occupations and compared them with workers in jobs that have little or no AI exposure.

The results show that unemployment rates have remained broadly similar between the two groups.

During the COVID-19 pandemic, workers in physical jobs experienced a much sharper unemployment spike. Since then, the gap between high-exposure and low-exposure occupations has remained relatively stable.

In other words, AI has not yet triggered a large wave of job losses.

Is Slower Hiring of Young Workers an Early Warning Sign?

While unemployment remains stable, the study finds suggestive evidence of another shift.

Hiring of younger workers into AI-exposed occupations appears to be slowing.

Researchers analysed job transitions among workers aged 22 to 25, tracking how often they moved into new jobs each month.

The results show that since 2024, entry into highly exposed occupations has declined slightly, while hiring into less exposed roles has remained stable.

The job-finding rate for young workers entering AI-exposed occupations has fallen by roughly 14 percent compared with 2022 levels.

Importantly, this slowdown affects hiring rather than layoffs. Many young workers may simply be entering different fields, staying in education or delaying career transitions.

Also read: India AI Summit 2026: Beyond the Chaos, India’s AI Moment Has Arrived

What Would a Real AI Job Shock Look Like?

The study outlines scenarios that would clearly signal AI-driven disruption.

For example, if unemployment among highly exposed workers doubled, similar to the surge during the 2008 global financial crisis, it would become visible in the data quickly.

Likewise, a scenario where all workers in the top 10 percent of exposure lost their jobs would raise overall unemployment dramatically.

However, current data show nothing close to such outcomes.

Instead, the labour market impact of AI so far resembles gradual adjustment rather than sudden shock.

What Does This Mean for India’s Job Market?

Although the research focuses on U.S. data, the patterns are highly relevant to India.

India’s workforce is heavily represented in sectors that overlap with AI-exposed occupations, including:

• IT services
• software development
• data processing
• customer support
• financial operations

India’s technology services industry employs over 5 million professionals, according to NASSCOM estimates. Many of these roles involve tasks that AI systems can assist with.

At the same time, the country also has a large workforce in manufacturing, logistics, agriculture and hospitality sectors where AI exposure remains relatively low.

This dual structure means India may experience uneven AI impacts across sectors, with white-collar roles adapting faster while physical jobs remain largely unaffected.

Also read: Meta Patents AI That Could Keep Users Posting After Death

Is AI Also Reshaping Content and Media Work?

Beyond traditional office roles, AI is already influencing content creation and media industries.

Generative tools are now assisting with:

• script writing
• video editing
• design layouts
• content research
• news summarisation

For digital creators, journalists and video producers, AI is increasingly becoming an assistive tool rather than a replacement.

In India’s fast-growing creator economy, AI tools are often used to speed up editing, subtitles, graphics and production workflows, allowing creators to produce content more efficiently.

This aligns with the broader trend identified in the study: AI is currently augmenting work more often than fully automating it.

What Lies Ahead in Tracking AI’s Real Impact?

The authors of the study emphasise that their work is only the first step in understanding AI’s labour market impact.

Future research will continue updating the exposure measures as AI capabilities expand and adoption spreads across industries.

Particular attention will likely focus on:

• new graduates entering AI-exposed fields
• changes in hiring patterns
• shifts in occupational demand

For now, the evidence suggests that AI is quietly reshaping workflows rather than abruptly replacing workers.

But as adoption accelerates and capabilities improve, the gap between what AI can do and what it actually does may narrow quickly.

When that happens, the labour market effects could become much more visible.