ECB: AI Already Cut High-Risk Jobs by 4% While Safe Jobs Grew 13%

Here is a number that should stop you mid-scroll: jobs most exposed to AI substitution have already shrunk by more than 4% in the United States since 2019, while the jobs AI can barely touch have grown by 13%. That is not a forecast about 2030. That is the European Central Bank measuring what already happened.

In its Economic Bulletin (Issue 4, 2026), the ECB looked at six years of US employment data and found a widening canyon between two kinds of work. The gap between high-substitution-risk and low-substitution-risk occupations now sits at roughly 15 percentage points — and it started widening fastest right after ChatGPT launched in late 2022. If you have wondered whether AI is "really" changing the job market or just generating headlines, this is the most sober answer yet: the change is real, it is measurable, and it is already in the data.

The 15-point wedge nobody voted on

Let's unpack what the ECB actually measured [Fact]. Economists at the central bank sorted US occupations by how exposed each one is to being substituted by AI, then tracked employment in those groups from 2019 through 2025. The result is stark. High-risk occupations didn't just grow more slowly — they declined outright, by over 4%. Low-risk occupations expanded by 13%. Stack those two trends against each other and you get a wedge of about 15 percentage points separating the winners from the losers.

That shift shows up in the structure of the whole labour market, not just at the margins. Low-risk jobs climbed from 23% to 25% of total US employment, while high-risk jobs slipped from 35% to 33% [Fact]. Two percentage points of the entire American workforce quietly migrated from exposed work to protected work in six years. That may sound small, but moving 2% of a 160-million-person labour force is millions of jobs changing shape.

Crucially, the ECB notes this divergence accelerated after ChatGPT's launch in late 2022 [Fact]. The timing matters. It is the difference between "automation has always reshaped work" and "this specific wave of generative AI is doing something measurable right now."

Who is on the wrong side of the line?

The ECB names names. On the high-risk side sit occupations like economists and graphic designers — knowledge and creative roles built around tasks that large language models and image generators now perform competently [Fact]. On the low-risk side sit electricians and high school teachers — jobs anchored in physical presence, hands-on judgement, and human relationships that AI cannot replicate.

If you work in a text-heavy or template-heavy profession, this pattern probably feels familiar already. Our own analysis flags several of these roles as highly exposed. Graphic designers, for instance, face intense pressure as generative image tools absorb routine production work — you can see the full breakdown on our graphic designers detail page. The same dynamic hits economists, whose forecasting and report-drafting tasks overlap heavily with what AI now drafts in seconds, and market research analysts, whose survey-synthesis work is increasingly automatable.

The protected side of the ECB's divide lines up just as neatly with the trades and in-person services. Electricians sit near the bottom of the substitution-risk spectrum because rewiring a building is stubbornly physical. High school teachers hold their ground because classroom management and adolescent mentorship resist automation, even as AI changes the tools they use.

The surprising part: your paycheck hasn't moved — yet

Here is the counter-intuitive finding that complicates the doom narrative. Despite the sharp split in how many jobs exist on each side, the ECB found no significant impact on wage growth from AI substitution risk since 2019 [Fact]. Workers in high-risk occupations are not — at least not yet — seeing their pay fall relative to everyone else.

How can employment crater while wages hold steady? The most likely explanation is that AI is reshaping the labour market through the hiring door, not the paycheck. Firms slow down new hires and let attrition shrink exposed roles, rather than cutting the pay of people already in those seats. That is why the damage shows up as fewer openings and weaker job growth long before it shows up in salary data. For a worker, this is the quiet danger: the job you have may feel fine, while the next job in your field is getting harder to find.

The ECB is careful here, and so should we be. Their framework does not directly control for AI adoption itself, and separate research raises an honest alternative: some of the decline in junior and entry-level hiring could reflect post-pandemic remote-work shifts rather than AI alone [Claim]. The honest read is that AI is clearly part of the story — the post-ChatGPT acceleration is hard to explain any other way — but it is not the only force bending these curves [Estimate].

Why does this matter so much more now than in past automation waves? Earlier waves of technology mostly displaced routine manual and clerical tasks, sparing the white-collar knowledge work that AI now targets directly [Claim]. The ECB's data captures the moment that exception ended. Economists, designers, and analysts spent decades assuming their cognitive work was a safe harbor. The 15-point wedge is the first hard evidence that the harbor is shrinking — and that the people most confident about their job security may be the ones with the least time to adapt.

What this means for your career

If you take one thing from the ECB's data, make it this: the AI labour shock is not evenly distributed, and it has already started. The relevant question is no longer "will AI affect jobs" but "which side of the 15-point wedge is my job on, and what do I do about it."

Three practical moves follow from the evidence. First, audit your task mix, not your job title. The ECB's risk scores track tasks — drafting, summarizing, generating — not job names. If most of your day is producing text, code, or images from a prompt-shaped brief, you are exposed regardless of your title. Second, lean into the human and physical edges the data rewards: judgement under ambiguity, in-person trust, hands-on problem solving, and the messy coordination that AI handles poorly. Third, watch the hiring door, not your current salary. Because the shock hits openings first, the early-warning sign in your field is fewer postings and longer searches — not a pay cut.

The hopeful part of this story is that it is a story about reallocation, not annihilation. Low-risk employment grew 13% — those are real, expanding opportunities, many of them within reach of a mid-career pivot. The workers who came out ahead in the ECB's data were not the ones with the safest job in 2019; they were the ones who moved toward work that AI complements rather than replaces. You can check where your own occupation falls on the exposure spectrum across our 1,016 occupation profiles, and treat the ECB's wedge as a map rather than a verdict.

Sources

• European Central Bank, "AI and the US Labour Market: Effects on Employment Growth," Economic Bulletin Issue 4/2026 (1 April 2026): ecb.europa.eu

This analysis was produced with AI assistance. Every statistic above is drawn directly from the European Central Bank's published Economic Bulletin and cross-checked against the original text. Figures are reported as the ECB stated them; our interpretation of what they mean for individual workers is our own.