Ask most people which workers face the greatest risk from AI and they will point to the lowest paid — the cashiers, the data entry clerks, the call centre agents. The data tells a more complicated story. When it comes to the new wave of large language model-driven AI, it is the highest earners — the lawyers, the software engineers, the financial analysts — who face the greatest total exposure. But here is the twist that changes the entire picture: exposure is not the same as risk. For high earners, AI exposure is predominantly a productivity multiplier. For low earners, what little AI exposure exists is more likely to displace than to augment. The chart below maps that distinction across five salary bands, and the pattern it reveals has significant implications for inequality, policy, and anyone planning a career in the decade ahead.
The chart ranks five income quintiles by share of workers exposed to AI automation risk, mixed or uncertain exposure, and augmentation potential. Use the toggles to isolate each exposure type. Hover over any bar for a breakdown of typical roles and the split between risk and opportunity within that band.
The headline finding nobody expected: AI exposure rises with salary
Every major study to examine this question arrives at the same counterintuitive conclusion. Exposure to AI is higher among people with higher levels of education who work in high-paying jobs, regardless of gender or race. This directly inverts the pattern of previous automation waves, where factory robots and software displaced low-wage, routine physical work. In contrast to both skill-biased and routine-biased technological change, AI susceptibility increases with wages — workers in the top income distribution face the highest AI exposure. The reason is structural: LLMs are fundamentally language and reasoning tools, and the jobs that rely most heavily on language and reasoning — law, finance, engineering, medicine — are also the best-paid ones.
But the type of exposure is what actually determines the outcome
Total exposure figures mask the distinction that matters most. AI’s effect on incomes is contingent on how it is used in workplaces — augmentative use of AI is positively correlated with higher wages, while automative use is negatively correlated. For workers in the top salary band, the dominant experience of AI is as a force multiplier: it writes first drafts, generates code, synthesises research, and handles the lower-value portions of complex tasks, leaving the high-judgment core of the role intact and more productive. For workers in the bottom salary band, where AI exposure is lower overall, the exposure that does exist is more likely to be automative — the task disappears rather than the task becoming easier.
The bottom quintile: less exposed, but more vulnerable when they are
Workers earning under £18k face lower overall AI exposure than any other group — but that relative shelter is partly misleading. Many roles in this band involve physical presence, manual dexterity, or interpersonal service that current AI cannot replicate: care work, food preparation, cleaning, maintenance. Nearly 50% of workers in the lowest income percentiles were exposed to traditional automation, compared to fewer than 20% of high-income workers — meaning this band has already absorbed significant displacement from earlier technology waves. The workers who remain are disproportionately in roles AI cannot yet reach. Those who are exposed, however — data entry clerks, basic customer service agents, routine admin workers — face displacement with limited augmentation benefit and fewer resources to reskill.
The middle bands: where the picture is most genuinely uncertain
Workers earning between £18k and £40k occupy the most uncertain territory on the chart. This is where the traditional “at-risk” roles cluster — customer service, administrative support, warehouse coordination, basic accounting — and where AI automation is already visibly reducing task demand. The distinction between codifiable and tacit knowledge suggests that AI may substitute for entry-level workers but augment the efforts of experienced workers — a pattern that is most sharply felt in the lower-middle band, where junior roles are disappearing faster than senior ones. At the same time, augmentation tools are beginning to reach this band through employer-provided software, creating a growing divide between workers who are learning to use these tools and those who are not.
Top earners: AI is amplifying the advantage
LLM exposure is highest for worker tasks at the upper end of annual wages, peaking at approximately $90,000 per year in the US. For the top salary quintile — lawyers, doctors, executives, senior software engineers — this exposure is overwhelmingly augmentative. AI-skilled workers saw an average 56% wage premium in 2024, double the 25% premium seen the previous year, with industries most exposed to AI seeing three times higher growth in revenue per employee compared to those least exposed. In practical terms, AI is enabling the highest earners to do more, bill more, and produce more — compressing the productivity gap between them and workers further down the income distribution. The inequality implications are significant and are already showing up in wage data.
What this means for the workforce overall
The picture that emerges is not one of mass unemployment, but of structural divergence. 35.9% of US workers reported using generative AI tools by December 2025, with adoption concentrated among younger, college-educated, and higher-earning employees. AI is not arriving uniformly across the labour market — it is concentrating its benefits at the top and its risks at the bottom, with a genuinely mixed and uncertain picture in between. The workers most at risk are not those who face the most AI exposure; they are those who face automation-type exposure with the fewest resources to respond to it. That distinction — between total exposure and the type of exposure — is the finding that every worker, employer, and policymaker should be taking seriously.
A note on methodology
This chart synthesises findings from six primary research sources: the IMF Gen-AI Staff Discussion Note (2024 and 2025 update), the Equitable Growth Working Paper on AI Exposure and Wages (October 2025), the Pew Research Center AI Exposure Study (2023), the Anthropic Economic Index (February 2025), the OpenAI and MIT study “GPTs are GPTs” by Eloundou et al. (2023), and the International AI Safety Report (2025). Salary bands use approximate UK equivalents; the underlying data is drawn from US-based research using Bureau of Labour Statistics occupational classifications. Figures represent estimates synthesised across multiple studies rather than a single audited dataset. The automation risk, mixed, and augmentation categories represent the balance of evidence across these sources at each income level rather than precise survey-derived percentages. This chart will be updated as new data becomes available.
5 thoughts on “AI Exposure by Salary Band”