Will AI take my job? What the studies actually say

Not for most people so far, but no study can promise anything about your own job. The big international studies agree on three points: AI could take over some tasks in between one job in four and almost four in ten worldwide, depending on the study; in many of those jobs it is more likely to change the work than to remove it; and up to mid-2026, job data showed no mass job losses. They stop at the future: they measure tasks and early trends, not what your employer will decide. And there is one early warning: in the United States, fewer young people are being hired in the jobs most exposed to AI.

Each study below was opened on its publisher’s site on 31 October 2026 and comes with its date and what it does not show. None is a forecast of how many jobs will disappear.

What is the difference between “exposed”, “replaced” and “helped”?

Economists see a job as a bundle of tasks. A bank clerk writes letters, checks documents, serves customers at the counter and calms down an angry one. A large language model, the program that writes the answers in ChatGPT, Gemini or Claude, can draft the letter. It cannot stand at the counter.

Exposed means AI could do some of the tasks in a job, at least in theory. It measures what is possible, not what will happen. Most of the big numbers in headlines are exposure numbers.
Helped (economists say “augmented” or “complemented”) means AI does part of the work and the person does the rest, often faster. The job stays, but it changes.
Replaced means fewer people are employed or hired to do the job. It is the only one of the three that costs jobs, and the hardest to measure, because hiring also falls for other reasons, such as high interest rates or a slow economy.

Exposed does not mean replaced. The International Monetary Fund’s example is judges: highly exposed, because AI can read and summarize legal texts, but protected, because “society is currently unlikely to delegate judicial rulings to unsupervised AI”.

What do the main studies say?

Study and date	What it measured	Main finding	What it does not show
International Labour Organization (ILO), the UN agency for work, Working Paper 140, 20 May 2025	Exposure: which of nearly 30,000 work tasks generative AI could do, applied to jobs worldwide	One worker in four is in a job with some exposure; 3.3% are in the most exposed group (4.7% of working women, 2.4% of men); clerical (office support) jobs are the most exposed	Any job actually lost. The ILO’s summary expects most jobs to be “transformed rather than made redundant”
International Monetary Fund (IMF), Staff Discussion Note 2024/001, 14 January 2024	Exposure to AI in general, plus how shielded each job is by its social, ethical and physical context	Almost 40% of jobs worldwide exposed: about 60% in advanced economies, 40% in emerging ones, 26% in low-income countries. In advanced economies about half of the exposed jobs may be harmed and the rest could gain	Jobs lost. The note calls its measures relative (a job is more or less exposed than others), and its views are the authors’, not necessarily the IMF’s
OECD, Employment Outlook 2026, July 2026	Real job data from Australia, Canada, the EU and the US	Companies “may have begun to automate certain jobs”, “although the data show little evidence of this to date”; the role of language models in young people’s job difficulties “remains limited”	Effects that may come later. The OECD says the evidence is “mixed and focusses mainly on the United States”
Stanford Digital Economy Lab, Canaries in the coal mine?, August 2025, revised 12 August 2026	Real payroll records of millions of US workers, up to June 2026	No sign of economy-wide job losses. Employment of 22 to 25 year olds in the most exposed jobs is 19% below where it would be had it kept pace with less exposed jobs; no such gap for experienced workers	Proof that AI is the cause: the authors call their results “early, descriptive indicators”. A working paper, not peer-reviewed; US only
Microsoft Research, Working with AI, July 2025, revised 23 October 2026	Real use: 200,000 anonymized conversations with Microsoft Bing Copilot, January to September 2024	People mostly ask the AI for help finding information and writing. The highest scores are in computer and mathematical jobs, office and administrative support, and sales	Effects on jobs or pay, which the authors say “will depend on hard-to-predict business decisions”. One product’s users only; Microsoft’s own product data. The revised version on arXiv is marked as a paper for the NeurIPS 2026 research conference
Anthropic Economic Index, reports of 15 January 2026 and 26 June 2026	Real use: samples of conversations with Claude, sorted by Claude itself, and a survey of about 9,700 Claude users	In November 2025, 52% of Claude.ai conversations were AI helping a person, 45% AI doing a task with little input. 10% of users surveyed thought losing their job in the next 12 months likely	The whole population: Anthropic writes that its survey “is not representative of the general population”. The maker’s own data
Banca d’Italia, QEF no. 878, October 2024, and QEF no. 1005, March 2026	Exposure of Italian workers (878); Italian firms with at least 50 employees (1005)	About 9 million Italian workers are highly exposed, 5 million of whom “could benefit from AI”. Firms that adopted AI shifted toward more skilled jobs “without yielding negative effects on overall employment”	Jobs lost (878 measures exposure); firms under 50 employees (1005), at a time when just over 10% of the firms surveyed used AI
Istat, Imprese e ICT 2025, 15 December 2025	How many Italian firms with at least 10 employees use AI	16.4% used at least one AI technology in 2025, up from 8.2% in 2024 and 5.0% in 2023	Anything about jobs gained or lost: it measures use, not effects

Two of these studies come from companies that sell AI assistants and draw on their own users’ conversations: Microsoft, maker of Copilot, and Anthropic, maker of Claude. This site is written partly with Claude, so keep in mind that the Anthropic Economic Index is Anthropic’s own data about its own product. Both describe how people use one tool, not the whole job market.

Has AI already cost jobs?

Not on a large scale, according to the data up to mid-2026, but there is an open question about young people.

The OECD looked at Australia, Canada, the European Union and the United States. Unemployment among young graduates, compared with the rest of the working-age population, had been rising “since well before the spread of generative AI models”, with no turning point when companies started using AI heavily, between mid-2023 and early 2024. Citing other studies, the OECD adds that the most exposed jobs sit in sectors such as finance and professional services, which cut hiring quickly when the economy worsens, so a drop there can look like an AI effect when it is not.

The Stanford study points the other way for one group. Young people in the most exposed US jobs fell well behind their peers, mainly because fewer were hired, and the drop was concentrated in jobs where AI is used to do the work instead of the person; where AI is used to help, employment was flat or rising. The authors stress that this does not prove AI is the cause.

The two findings can both be true. Stanford looks closely at one narrow group in one country; the OECD looks at young people entering the job market, with and without a degree, across several countries.

What do the studies say about Italy?

Banca d’Italia’s 2024 study found that the most exposed jobs in Italy are mainly in services, which employ many women and skilled workers. It also found that few people move between jobs with different levels of exposure, and that those who leave highly exposed jobs where AI could replace them mostly move to less exposed jobs “where wages tend to be lower”. Its March 2026 study adds that the companies that adopted AI also raised productivity and profits.

Istat’s figures show how early this still is. In 2025, 16.4% of Italian firms with at least 10 employees used some form of AI, double the year before, and among firms that considered AI but did not invest, almost 60% said the lack of adequate skills held them back.

Which jobs are most and least exposed?

The studies agree on the broad picture:

Most exposed: clerical work (ILO, IMF); computer and mathematical jobs, office and administrative support, and sales (Microsoft), whose researchers also name interpreters and translators, and writers and authors. The IMF’s example of a job where AI is more likely to replace than to help is telemarketing.
Exposed, but more likely helped: the IMF names surgeons, lawyers and judges, jobs with “a high degree of responsibility and interpersonal interactions”.
Least exposed: jobs done with the hands and in person. Microsoft found the lowest scores in jobs “that require physically working with people”, and the IMF gives dishwashers and performers as examples.

These studies are about software that reads and writes. Robots that do physical work are a separate and slower story, explained in What is embodied AI?.

What can you do about your own job?

List your tasks. Write down what you do in a normal week, not your job title.
Mark each task. Tasks on a screen with text, numbers or standard replies are the exposed ones. Tasks done with your hands, face to face, or where you carry the responsibility are the less exposed ones.
Use AI on one exposed task. Try an assistant on a draft or a summary and check every line. Keep confidential work data out of it: what you type is stored on the maker’s computers, and some makers, OpenAI among them, use it to train their models unless you turn that off.
Learn the skills. The IMF warns that “workers without AI-related skills risk reduced employment” even in jobs where AI mostly helps. Elements of AI is a free online course created by the University of Helsinki and MinnaLearn; the Italian edition is promoted by Fondazione Cotec with Roma Tre University.
Ask your employer. Under the EU AI Act, companies that use AI systems must take measures to support the AI literacy of their staff, according to the European Commission’s questions and answers, updated 20 October 2026. It is a fair reason to ask for training.

Where to go next

What is an AI agent?: the AI that clicks and types for you, doing tasks instead of helping with them.
Is ChatGPT free?: what the free version does and when a paid plan makes sense, if you want to try AI on your own tasks.
Generative AI: the technology most of these studies measure.

Common questions

Is it true that AI will replace 40% of jobs?

No study opened for this page says that. The 40% comes from an IMF note of January 2024, and it counts jobs exposed to AI, meaning jobs where AI could do some of the tasks. In rich countries the IMF estimated that about half of the exposed jobs could be harmed and the other half could gain from AI. It is an estimate of possibility, not a count of jobs lost.

Will AI replace lawyers?

The IMF puts lawyers, surgeons and judges among the jobs that are highly exposed to AI but also well protected, because of their responsibility and their contact with people, so AI is more likely to help them than to replace them. The same note warns that workers in these jobs who lack AI skills can still lose out. None of the studies on this page counts lawyers' jobs lost to AI.

Which jobs can AI not replace?

No study promises that any job is safe forever. Today the least exposed jobs are those done with the hands and in person: Microsoft's researchers found the lowest scores in jobs that require physically working with people, and the IMF gives dishwashers and performers as examples of low exposure. These studies are about software such as chatbots, not robots.

Are young people losing jobs because of AI?

The evidence is mixed. Stanford researchers found that in the United States employment of 22 to 25 year olds in the most AI-exposed jobs was 19% below where it would have been by June 2026, mainly because fewer were hired, but they call this an early sign, not proof. The OECD says unemployment among young graduates was rising, compared with other adults, well before generative AI spread, and that the role of AI in it appears limited so far.

Where can I learn to use AI for free?

Elements of AI is a free online course created by the University of Helsinki and MinnaLearn; in Italy it is promoted by Fondazione Cotec with Roma Tre University. If your employer uses AI tools, ask about training: under the EU AI Act, companies that use AI systems must take measures to support the AI literacy of their staff.

Topics: AI and jobs, work, studies, generative AI, Italy