"AI will cost 300 million people their jobs." That sounds like a forecast. Goldman Sachs first published the widely cited figure on April 5, 2023. It referred to the equivalent of 300 million full-time jobs worldwide potentially exposed to automation by generative AI. This is an exposure estimate, not a forecast that 300 million people will be laid off. The analysis says AI is more likely to complement many tasks, and that not every automated task will translate into layoffs. In a separate US base case from March 2026, Goldman estimates that 6% to 7% of workers could be displaced during an adoption period of about ten years. Goldman Sachs on the 2023 estimate and its 2026 US scenario.
An analysis published in September 2026 examined more than 800 occupations. It found that 10% occupational AI exposure was associated with 0.1 percentage point lower annual headcount growth in France, Canada, and the US. This observed association does not prove causation, count people laid off worldwide, or measure completed job cuts. Goldman Sachs on the September 2026 international labor-market analysis.
Which figures are forecasts, which track actual employment, and which only count announced cuts? Those distinctions determine what you can conclude from a statistic.
- The WEF projects a net 78 million new jobs by 2030 across several global trends. It is not an AI-only forecast.
- Challenger counted 120,136 announced US job-cut plans citing AI from January through September 2026. That is not a verified count of people who have already lost jobs.
- Stanford and ADP found a 19% employment gap for 22- to 25-year-olds in highly AI-exposed occupations. The researchers call it an early descriptive finding, not proof of causation.
1. The Key Numbers at a Glance
2. WEF Forecast to 2030: 170 Million New Jobs, 92 Million Displaced
The World Economic Forum's Future of Jobs Report 2025 is based on 1,043 employer responses across 55 economies. Those companies represent more than 14.1 million employees. For its global estimates, the report covers selected occupations representing 1.18 billion workers. Its forecast combines several trends, including technology, demographics, the green transition, and economic uncertainty. WEF methodology and sample
The 170 million projected new and 92 million displaced jobs add up to a net forecast of 78 million by 2030. This is not an AI-only projection. For AI and information processing combined, the WEF expects 11 million jobs to emerge and 9 million to be displaced. Employers also expect 39% of core skills to change by 2030. In response to AI, 77% plan to upskill their workforce, while 41% expect to downsize where AI can replicate roles. These plans can coexist. WEF skills outlook and workforce strategies.
The WEF links job forecasts to multiple trends. Technology is one factor, alongside demographics and economic change. This list is not a ranking of AI risk alone:
3. What Employers Have Actually Announced
Challenger, Gray & Christmas tracks announced US job cuts and records the reasons given by employers. Its figures are announced plans and stated attributions, not independently verified actual layoffs:
Source: Challenger, Gray & Christmas, September 2026
From January through September 2026, employers cited AI in 120,136 announced cuts. Challenger says that was about 21% of all announced cuts. September accounted for 3,961, or about 9% of the month's total. AI ranked fifth among stated reasons that month. It had led for five months in a row from March through July, then dropped to fourth place in August. The last cumulative total Challenger published since it began tracking AI in 2023 was 184,538 through July 2026. That is a historical subtotal, not a cumulative figure through September. July report with cumulative total and September report.
Even an employer-stated AI reason does not show how many people actually lost jobs or what share of a reduction was caused by AI. Conversely, making work more efficient with AI does not automatically mean roles were cut. Announced layoffs, internal moves, unfilled vacancies, and completed tasks are different measures.
4. Young Workers in AI-Exposed US Occupations
A study by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen at Stanford University analyzes high-frequency ADP payroll records covering millions of US workers. Its revised analysis includes data through June 2026. The researchers find no widespread displacement across the US economy, but they do find a marked gap for younger workers.
The result in one sentence:
In June 2026, employment among 22- to 25-year-olds in highly AI-exposed occupations was 19% below the comparison path for similarly aged workers in less-exposed roles. The researchers say the gap appears to be driven mainly by lower hiring of young workers, not higher layoffs. They describe this as an early, descriptive indicator, not causal proof that AI alone created the difference.
The newer ADP monthly report adds an August 2026 comparison. Year over year, employment among 22- to 25-year-olds in highly AI-exposed jobs fell 4.4%. In the least-exposed jobs, it fell 2.0%. Employment for all workers in that age group fell 3.1%. This is a year-over-year comparison and a different measure from the 19% gap against a comparison path in the Stanford study. The researchers also note that the gaps are more pronounced in the ADP sample than in national comparison data. Stanford study and methodology and ADP August data.
PwC offers a different, complementary measure. Among 2.4 million US entry-level job postings it analyzed, postings for highly AI-exposed entry-level roles grew 35% from 2019 to 2025, while other entry-level roles declined 10%. The exposed roles were also seven times more likely to ask for skills traditionally associated with senior jobs, such as judgment or leadership. Job postings are not counts of completed hires or employed people. PwC AI Jobs Barometer 2026.
5. Automation or Augmentation? What Usage Data Shows
Anthropic's Economic Index offers a look at how its own product is used. From a random sample of one million Claude.ai conversations collected from November 13 to 20, 2025, it classified 52% as augmentation, 45% as automation, and 3% as neither. The classification distinguishes people iterating with Claude from task completion with little back-and-forth:
Does a person work through several rounds with Claude, or delegate a task with little back-and-forth?
The chart describes Claude.ai conversations, not every Claude product, worker, or the labor market as a whole. Anthropic's report, published June 26, includes a survey of about 9,700 Claude users. Anthropic linked their responses to randomly sampled usage sessions from mid-May through early June 2026 across Claude.ai, Cowork, and Claude Code. The company says the survey is not representative of the general population. More than a third considered a substantial change in responsibilities likely for themselves, a peer, a junior colleague, or a senior colleague over the next year. 10% considered losing their own job likely or very likely. More than a third estimated a probability above 60% that a junior colleague would involuntarily lose a job in the next year. These are expectations from a selected user group, not observed layoffs. Anthropic Economic Index, June 2026.
6. Wage Premiums for Jobs Requiring AI Skills
PwC's Global AI Jobs Barometer analyzed more than one billion job postings from 27 countries. The three values in the chart refer to observation years, not report publication years:
For 2025, PwC estimates an average wage premium of 62% in postings for jobs requiring AI skills compared with similar postings without that requirement. The report puts the 2024 figure at 57%. Values vary widely by sector. A premium in job postings does not mean every worker earns that much more after taking a course or earning a certificate.
PwC found that job postings requiring specific AI skills grew 69% in 2025, compared with 9% for the overall market. Companies in the most AI-exposed sectors reported 52% headcount growth from 2018 to 2025, compared with 36% in the least-exposed sectors. Productivity growth was 34% versus 24%. These are associations in PwC's data, not proof that AI alone caused the growth. PwC AI Jobs Barometer 2026.
7. Occupational Forecasts and Demand for AI Skills
The WEF lists big-data specialists, fintech engineers, and AI and machine-learning specialists among its fastest-growing occupations through 2030. These are employer expectations, not a count of jobs created solely by AI. A separate measure shows how frequently job postings mentioned AI skills:
Singapore led in 2025 at 4.69%. The chart adds the next four countries with the highest shares and the US. Canada, at 3.00%, was still ahead of the US at 2.56%. Germany is not among these six selected countries. The measure counts job postings requiring AI skills, not the number of people in AI occupations or newly created jobs. Stanford AI Index 2026, chapter 4.
8. Germany: Different Measures, Different Answers
Several credible German surveys now exist. They measure different things and deliberately belong side by side, not merged:
The Bitkom survey published in March 2026 was fielded in calendar weeks 2 to 6 and covered 604 companies with at least 20 employees. 41% already used AI. The 19% that reported cutting jobs as a result of AI came from the subgroup of 249 companies already using AI. Bitkom March results presentation on AI use and its March press release.
A separate Bitkom IT labor-market study, published on September 3, 2026, is based on two representative surveys of 853 and 602 companies with at least three employees. They were fielded in calendar weeks 18 to 25. In it, 10% of companies using AI in their IT departments said they had occasionally cut IT jobs. None reported more frequent cuts; 85% said they had not yet cut IT positions because of AI. Bitkom IT labor-market study.
Another representative Bitkom survey of 603 companies with at least 20 employees was conducted by phone in calendar weeks 28 to 33 and published on September 14. 57% were already using AI, compared with 36% the year before and 20% two years earlier. Over the next two years, 45% expected a slight and 12% a substantial decline in total headcount due to AI. 14% expected a slight increase, no company expected a substantial increase, and 29% could or would not answer. Among companies using AI, only 9% expected a substantial decrease, compared with 15% of companies not using AI. AI-using companies were more likely to expect an increase, 18% versus 9%. These are expectations and group differences, not proof that using AI itself caused the difference. The surveys ask different questions of different groups and are not directly comparable. Bitkom survey on AI use and employment.
The 2025 ifo survey asked companies already using or planning to use AI. 27.1% expected job cuts within five years; 5.2% expected job growth. These are forecasts, not completed changes. From 2022 data, IAB estimated that 38% of employees covered by social insurance worked in occupations where at least 70% of tasks were technically substitutable. That is a potential across computer technologies, not measured AI replacement. ifo employment expectations and IAB substitution potential.
A further ifo survey of more than 3,000 companies already using AI asked about wages over the next five years. 53.3% expected lower pay for workers without a university degree and under five years of experience. For workers with a university degree, 26% expected a positive wage effect. These are employer expectations, not observed wage changes. ifo wage survey. For more on AI adoption in companies, see my AI agent statistics.
9. What Workers Think
Pew surveyed 3,488 US adults from June 22 to 28, 2026, publishing the results in August. 71% expected AI to lead to fewer US jobs over the next 20 years, up from 64% in 2024. An international Pew survey fielded 42,151 people across 36 countries from February 8 to May 13, with separate US samples. In 34 of the 37 countries studied, more people expected job losses than job gains from AI. These figures measure public expectations, not employment outcomes. Pew US survey and Pew international survey.
Those survey numbers describe expectations. To understand what is happening to employment, payroll data such as Stanford and ADP's is more informative, though it also cannot prove the cause on its own.
10. Conclusion: Disruption Yes, Wipeout No
The data do not show a simple wave of AI layoffs. Challenger counted 120,136 US job-cut plans citing AI through September, while Stanford and ADP found no widespread displacement in their US payroll sample. At the same time, employment among young people in highly exposed occupations sat well below its comparison path. How much AI itself caused that gap cannot yet be quantified with confidence.
Wage premiums, job postings, and employer forecasts are no guarantee for any one worker. They show where demand is shifting and which groups deserve attention. The essential distinction remains simple: exposure is not a layoff, a forecast is not an outcome, and an employer-stated reason is not proof of causation.
For how many people use AI in the first place, see my article on how many people use AI. And I dig deeper into the agent productivity debate in my AI agent statistics.
