AI and jobs: the statistics, with sources

Goldman Sachs estimates that 300 million full-time jobs globally are exposed to automation by generative AI.

Exposure is not replacement. The estimate counts jobs where a meaningful share of tasks could be automated, which is the population from which actual displacement is drawn.

The 2013 Oxford study by Frey and Osborne estimated that 47% of US jobs sat in the high-risk category for computerisation over the following one to two decades.

The most misquoted number in the field. It described susceptibility over 10-20 years, not a forecast that 47% of jobs would disappear - and it predates large language models entirely, so it understates exposure for knowledge work and overstates it for physical work.

The OECD finds 14% of jobs across member countries face high automation risk, defined as 70% or more of tasks being automatable, with a further 32% facing significant task change.

The OECD's task-level method is the closer analogue to how AI actually arrives at work: most people keep their job title while the contents of the job change underneath them.

Brookings finds that AI exposure concentrates in better-paid, better-educated white-collar occupations, reversing the pattern of earlier automation waves.

Previous automation moved up from the factory floor. This wave started in the office, which is why the roles that felt safest in 2015 are the ones asking the question now.

The World Economic Forum projects that 85 million jobs will be displaced by the shift in the division of labour between humans and machines.

The number is routinely quoted without its other half: the same report projects 97 million new roles emerging. The constraint is not the count of jobs, it is whether workers can move between them.

The World Economic Forum projects 97 million new roles will emerge from the same human-machine transition that displaces 85 million.

Net job creation is positive on paper. It is negative for any individual whose current skills do not transfer, which is why role-level exposure matters more than the aggregate.

Job Security Meter has tracked 17 frontier AI model releases during 2026, of which 9 were flagship launches, through August 2026.

The pace, not any single model, is the thing that breaks career planning. A skill assessed as safe against one year's frontier is being assessed against a different frontier by the next quarter.

Three frontier models - Claude Opus 4.7, GPT-5.5 and DeepSeek V4 - launched within eight days of each other in April 2026, the tightest cluster of frontier releases recorded.

Each removed a different objection to automating knowledge work: complexity, step count, and cost. A career defence resting on any one of them expired inside a single week.

Across the 36 occupations in the Job Security Meter Role Resilience Index, the median substitution probability is 41%.

Substitution probability estimates how much of a role's task mix current AI can perform, assessed task by task rather than by job title. The median role is therefore substantially exposed without being replaceable.

39% of the 36 occupations audited by Job Security Meter have a substitution probability of 50% or higher, meaning current AI can perform at least half their task mix.

Crossing 50% rarely eliminates a role. It changes what the role is for: the remaining half becomes the whole job, and it is usually the judgement, relationship and accountability half.

Accountant is the most exposed occupation in the Job Security Meter Role Resilience Index, with a substitution probability of 92%.

Traditional bookkeeping is at extreme risk, while forensic accounting and tax strategy remain strong.

Nurse / Healthcare Worker is the least exposed occupation in the Job Security Meter Role Resilience Index, with a substitution probability of 12%.

Nursing is one of the most AI-resilient professions in the world. While AI handles documentation and monitoring, the physical care, emotional empathy, and split-second clinical judgment nurses provide cannot be delegated to a machine.

9 of the 36 occupations audited by Job Security Meter carry a High automation risk rating, across 12 industry categories.

High risk means the majority of the role's current task mix is reachable by generally-available AI today, not that the occupation disappears on a known date.