Goldman Sachs says AI could displace up to 9% of US workers, faster in a recession

Goldman puts a number on AI job displacement

Goldman Sachs economists say artificial intelligence could displace up to 9% of US workers as it spreads through the economy, roughly one in 11 people in work today. The estimate comes from a research note by economist Joseph Briggs published on Thursday 22 October 2026, as reported by Fortune and Business Insider. Goldman's research notes go to clients and are not public, so the figures here come from those reports and from Briggs's earlier public comments.

The note adds that slower hiring could weigh on employment by a further 3% of workers. Its most striking finding is about timing rather than size. Looking at decades of data, Briggs found that job losses from automation stay limited while the economy is healthy and accelerate in downturns. "A more concerning takeaway from our analysis is that a cyclical slowdown would likely accelerate AI-related automation and job losses, thereby amplifying the broader business cycle," the note says.

Business Insider attributes parts of the same report to Goldman's chief economist, Jan Hatzius, while Fortune names Briggs as the author. Both describe the same baseline: AI eventually lifts productivity by 15% in the US and other developed economies.

How Goldman gets to 9%

The 9% figure is a calculation rather than a survey. Goldman looked at how past technology-driven productivity gains affected jobs. According to Fortune, each 1% gain historically pushed the unemployment rate up by 0.3 to 0.4 percentage points at its peak, an effect that faded to zero after about two years. Over the same period, it raised the cumulative rate of gross job losses by 0.5 to 0.6 points. Multiply the top of that range by the assumed 15% productivity gain and the result is 9%.

That distinction matters. The note measures gross job losses, meaning people pushed out of their current roles, not the number who end up unemployed. "Some workers that are displaced may find jobs relatively quickly, while others may decide to exit the labor force," Briggs writes. In a Goldman Sachs Exchanges podcast published in July, he said that 9% would correspond to about 15 million workers over a period of about ten years, and that even then "the unemployment rate increase in any given year would be less than one percentage point."

Goldman also expects new jobs to absorb displaced workers over time. On the podcast, Briggs noted that around 85% of US job growth over the past 80 years came from new kinds of positions created by technology, and that the US creates around 30 million jobs a year while destroying about 29 million. Even a 5% faster pace of job creation, he said, would be enough to reabsorb the workers he expects AI to displace.

Why a recession could speed it up

The new part of the analysis is the link to the business cycle. Briggs tracked the share of Americans in routine jobs such as clerks, secretaries, machine operators and mechanics. That share "has remained stable during expansions but declined sharply during downturns," he writes, dropping in steps in the recessions of the early 1980s, 2001, 2008 and 2020 and never recovering.

The pattern held industry by industry. Goldman found the rise in unemployment from a technology-driven productivity gain was 0.1 to 0.2 percentage points larger when the overall jobless rate was rising. Across 25 occupations that past technology upended, employment fell an extra 4% over five years for every percentage point increase in the national unemployment rate.

The explanation is familiar to economists: companies use slowdowns to restructure and cut costs they have tolerated in good times. "Historically automation-related labor disruption is quite cyclical, with companies tending to automate work and shed labor costs when the economy starts to slow," Briggs writes. Some of that is already visible in individual firms. This week Barclays named its first chief AI officer while cutting 5,000 jobs, although the bank is not in a downturn.

What history says about jobs hit by technology

To gauge how severe things could get in a single line of work, Goldman studied 25 occupations hit hard by earlier technology, including photo processors, movie projectionists, proofreaders, file clerks and bank tellers. On average, employment in those jobs fell 25% in the five years after it started to decline, "suggesting that very large job losses are possible when technology makes labor redundant." Briggs calls those numbers "likely upper bounds" for AI, because he picked the cases where technology hit hardest.

Most of the shrinkage came from people losing or leaving jobs rather than employers hiring fewer people. Separation rates rose in 19 of the 25 occupations, by 3.8 percentage points on average, while hiring rates fell in 13, by 0.7 points on average. That is why Briggs says AI-related layoffs "should provide a good real-time guide" to disruption.

The note also lists reasons the 9% could be too low. Displacement ran higher during the late 1990s and early 2000s productivity boom, which Goldman calls the period most comparable to today, and after unusually large productivity jumps. Goldman expects AI's boost to be bigger than most past ones.

Why the impact looks small so far

The note opens by acknowledging that fear has faded. "Concerns of imminent AI-related labor displacement have cooled in recent months after hitting a fever pitch this spring," Briggs writes. Layoffs attributed to AI in Challenger, Gray and Christmas's tracking have fallen sharply from their spring peak, Fortune reports, and news coverage of AI job losses is back to levels seen at the start of 2026.

Job growth in the industries most exposed to AI has run about 10,000 a month below its long-run trend over the past three months, roughly offset by hiring to build AI infrastructure. Business Insider notes that AI was still the top cited reason for announced job cuts in September at 21%. Company data points in the same direction: an Anthropic study this month found agent hours outnumbering staff hours and junior technical hiring falling.

Goldman gives two reasons the overall effect is still small. Only 23% of US establishments have built AI into regular workflows, so most productivity gains are still ahead. And with the economy and labour market resilient in 2026, "companies may have had less incentive to focus on aggressively cutting labor costs." Under Goldman's baseline of near-potential growth next year, it expects AI disruption to rise gradually rather than arrive as an abrupt shock.

How the forecast has changed

Goldman has been raising its number. In August 2025, Briggs and economist Sarah Dong estimated that AI could displace 6% to 7% of the US workforce if widely adopted, within a range of 3% to 14%, using the same 15% productivity assumption. By the summer of 2026, Briggs was putting the figure at about 9%. On the podcast he described those workers as being "reallocated to new positions"; the new note calls them displaced.

The language shift reflects a change in emphasis rather than a contradiction. Goldman's US economists said in October 2025 that they were skeptical of the boldest claims that AI would cause very high unemployment, and Briggs still expects displacement to be temporary. What the new note adds is a warning that the pain could be concentrated in the next recession instead of spread smoothly across a decade, which would make it far more visible.

What to watch

The first signal to watch is AI-attributed layoffs, which Goldman itself calls a real-time guide. The second is the wider economy: Fortune notes that the AI layoff panic peaked in the spring, and any slowdown would, on Goldman's reading, bring automation decisions forward. Governments are starting to plan for that. China says it will screen AI projects for job losses, and the UK's Budget AI package includes £10 million a year to study AI's effect on jobs.

The usual caveats apply. The 9% figure depends on a 15% productivity assumption that is itself a forecast, and on the past relationship between technology and jobs holding for a very different technology. It counts people displaced from roles, not permanent job losses. But it gives policymakers and companies a concrete number to plan around, and a reason to think the timing of the next downturn may matter as much as the pace of AI progress.