Is AI Really Taking Jobs? What the 2026 Data Actually Shows

TL;DR. The "AI is gutting white-collar jobs" narrative isn't showing up in the spring 2026 aggregate data — openings are near a two-year high, hiring is broad-based, and AI-exposed occupations don't have higher unemployment. The real, measurable exception is the bottom rung: entry-level and recent-grad hiring is softening. Not "AI is fine," not "the jobs apocalypse is here" — the honest middle.

Few AI questions get more emotional than "is it going to take my job?" — and few get worse answers, because both sides have an incentive to exaggerate. The doom version sells clicks; the boosters wave it away. As of the spring 2026 labor reports, here's what the actual numbers say.

The aggregate picture: not yet

If AI were broadly destroying jobs, you'd expect it in the headline labor data. It isn't there — at least not at the level of the whole economy.

• Openings are near a two-year high. US job openings climbed to roughly 7.6 million in April 2026, the most in nearly two years (Bureau of Labor Statistics JOLTS data, reported by Fortune).
• Hiring was broad-based. ADP's May report showed about 122,000 private-sector jobs added, with gains in eight of ten sectors — "more broad-based than we've seen in the last few years," per ADP's chief economist.
• AI-exposed jobs aren't doing worse. This is the counterintuitive one: MIT Technology Review's analysis of BLS data found the unemployment rate for the occupations most exposed to AI is lower than for less-exposed ones — and no sign of people fleeing "threatened" white-collar work for manual jobs.

The blunt summary from MIT Technology Review: despite the growing hysteria, there's still scant evidence AI has had a large-scale impact on the labor market.

The real exception: the first rung is wobbling

Here's the part the reassuring takes skip — and it's the part worth taking seriously. Aggregate stability can hide a shift underneath, and the shift is concentrated where careers start.

• A Stanford Digital Economy Lab working paper (November 2025) found that workers aged 22–25 in the most AI-exposed occupations experienced roughly a 16% relative decline in employment after generative AI spread — even after controlling for other factors that affect hiring.
• Unemployment for recent college graduates has run around 5.6% — well above the rate for all workers, a level not seen since the pandemic and the years after 2008.
• The Information sector (software publishing, data processing, telecom) shed about 9,000 jobs in May even as most sectors grew.

The intuition fits: the tasks AI does best right now — drafting, summarizing, basic research, first-pass code — overlap heavily with what junior employees were hired to do. If a senior person plus AI can absorb that work, the entry-level opening quietly doesn't get posted. That's not a mass layoff; it's a missing rung, and it's harder to see.

Why both things are true at once

It feels contradictory — "AI hasn't broadly cut jobs" and "AI is hitting early-career hiring." Both hold, for understandable reasons:

• Layoffs are loud; non-hiring is silent. A company announcing cuts and citing AI makes news. A team that simply never opens the junior req makes none — but it's the same lost job.
• The data lags. Labor statistics trail reality by months. Early, narrow effects may not yet show in the aggregates.
• Augmentation ≠ replacement (so far). The dominant 2026 pattern is AI making existing workers faster, not wholesale replacing roles. That can still reshape who gets hired — fewer entry slots, more demand for people who can direct AI — without moving the top-line unemployment number.

One honest caveat in both directions: this is early, the measurement is imperfect, and the picture can change quickly. Treat anyone claiming certainty — utopian or apocalyptic — with suspicion.

Update, June 4: what major employers are doing about it

Fresh this week: While aggregate labor statistics remain stable, some of the loudest cuts of 2026 are being named and framed explicitly around AI — and at least one CEO is trying to turn that into a replicable model.

Verizon's CEO Dan Schulman — who took over the telecom giant from Hans Vestberg in October 2025 — announced 13,000 layoffs, the company's largest single headcount reduction: roughly 13% of total employees and 20% of non-union management roles. The company's stated target is $5 billion in operating-expense savings by end of 2026, with Schulman telling investors Verizon expects to be "substantially complete with that entire AI tech stack by July."

Schulman's public stance on the broader picture is more candid than most Fortune 10 executives. He's warned publicly that US unemployment could reach 20% to 30% within two to five years as AI handles tasks currently performed by humans. He isn't framing this as distant — he describes it as the direction things are already moving, and he uses his own company's cuts as evidence.

The concrete response he's building around it: a $20 million reskilling and career-transition fund, set up specifically to help displaced workers develop skills for the AI-era labor market. More notably, Schulman is now lobbying other Fortune 100 CEOs to replicate the model — essentially arguing that companies cutting for AI have an obligation to fund the transition.

How to hold this alongside the aggregate data: Schulman's 20–30% prediction is a forecast, not current reality — and the MIT Technology Review data cited earlier in this article is about today's numbers, not the next five years. These two things can both be true: aggregate statistics are still stable, and a major company's CEO is positioning as if significant disruption is coming and cutting accordingly. The question for any individual professional isn't which view "wins" — it's whether you're building skills that hold value in either scenario.

June 12 update: What Americans actually fear

The labor data above measures outcomes. Two surveys released this week measure something different: public perception. The gap between the two is worth understanding.

Reuters/Ipsos poll (June 3–8, 2026 · n = 4,531)

• 53% of Americans worry AI will cause someone in their household to lose a job
• 71% worry AI will put too many people out of work permanently
• 73% are concerned about increasing AI use broadly — up from 68% in a comparable poll in 2023

Concern is consistent across age, gender, and education levels. Democrats (61%) express more worry than Republicans (47%); independents fall in the middle at 51%.

Anthropic/YouGov "Public Record" survey (Nov–Dec 2025 · n = 51,993 Americans)

Anthropic's large-scale national poll — conducted with YouGov across all 50 states and released this month — adds a second fear to the picture:

• 64% worry AI will cost them their jobs (close to the Reuters/Ipsos figure)
• 56% fear cognitive dependency — Anthropic's term for the concern that AI use will erode the ability to think independently
• Only 15% trust AI companies to responsibly guide the technology's development; independent experts ranked far higher, at 43%
• 70%+ support government oversight of AI, spanning party lines

The cognitive dependency finding is the new element here. The worry that AI might replace your job has been tracked for years. The worry that AI might replace your judgment — that it will gradually do your thinking for you if you let it — is newer in public data and more personally felt. It's not about a discrete event (getting laid off) but about a gradual erosion.

What the gap between perception and data means

Fear consistently outpaces realized disruption in technological transitions. The 53% who fear job loss aren't wrong about the direction of change — they're responding to genuine uncertainty and loud signals. But fear at scale has its own consequences: it shapes training investments people make, how companies communicate about AI, and how quickly regulatory pressure builds, independent of what labor statistics show.

The cognitive independence concern is harder to dismiss than the jobs fear. No long-run data exists yet on what sustained AI use does to unaided reasoning at scale. The constructive response is to use AI deliberately — as a tool you direct and critique, not a voice you defer to. The skill of critically evaluating AI output, catching its confident errors, and knowing when not to trust it is the one that keeps your judgment intact even as the tools get more capable.

June 14 update: Two things that complicate the picture further

Two new findings from June 14 add important layers to everything above.

The hidden jobs gap — most displaced workers never file unemployment claims

Unemployment claims have looked stable at 200,000–250,000 per week. But a Fortune investigation published today reveals why that number may badly undercount what's actually happening: according to a Bureau of Labor Statistics survey, roughly 75% of unemployed Americans never apply for unemployment benefits at all.

The most common reason: 55% of non-filers assumed they weren't eligible, even when they were. Another 17% expected to find a new job quickly and didn't bother. Only about 55% of those who do apply end up receiving benefits. The system was designed to replace 50% of wages; in many states it now delivers closer to 30% — and some states (Arkansas, Florida, North Carolina) have cut benefit duration from the original 26 weeks to just 12.

The practical implication for reading labor data: if the historical benefits take-up rate persists during AI-driven layoffs — and there's no reason to expect it to improve in a workforce less unionized than it has ever been (union membership hit a historic low of 9.9% in 2024, and union members are twice as likely to file) — then weekly jobless claims are a significantly lagging, understated signal. The "stable" headline number and the "120,000 tech workers laid off in 2026" number can both be true at once, for structural reasons that predate AI.

As Columbia professor Alexander Hertel-Fernandez put it: "Really, a wholesale reform is needed, especially as we think about…AI." The safety net was built for an economy that doesn't exist anymore.

AI currently costs more than the workers it's replacing

Here's the counterintuitive data point that explains why the disruption isn't showing up faster: for many organizations right now, AI compute costs more than the employees it replaces.

Bryan Catanzaro, Nvidia's Vice President of Applied Deep Learning — one of the people who should know — told Axios in April 2026: "For my team, the cost of compute is far beyond the costs of the employees." A 2024 MIT study reached a similar conclusion: AI automation is currently economically viable in only about 23% of the roles examined; in the other 77%, human workers remain cheaper.

That hasn't stopped Big Tech from committing roughly $740 billion in AI capex in 2026 — a 69% jump from 2025. The bet is temporal: inference costs for large models are expected to fall roughly 90% by 2030 (Gartner). Companies cutting workers now are positioning for an economics flip that hasn't happened yet.

What this means for how to read the disruption: companies making AI-driven cuts today are making a long-run investment, not reflecting current efficiency. The displacement is real and deliberate; it's just not driven by AI being cheap yet. If inference costs do fall 90% this decade, the economic case for automation will spread from the 23% of roles where it already pencils out to a much wider share — and the structural non-filing problem in unemployment insurance will make it harder to see in the data until it's well underway.

June 29 update: The economic case for optimism — and why entry-level workers still bear the cost

Two analyses published today in Fortune add important angles not yet covered in this post.

The Jevons Paradox: AI probably creates more total professional work — just not at the entry rung

The oldest counterargument to "AI will eliminate professions" is getting renewed attention. Economist William Stanley Jevons observed in 1865 that making steam engines more efficient didn't reduce coal consumption — it increased it, because lower cost unlocked demand that hadn't previously existed. The principle has held repeatedly in professional services: accounting software gutted bookkeeping jobs while the CPA profession expanded. Legal research tools reduced hours per matter while firms took on more clients. Apollo Global's chief economist Torsten Slok called this the most important counterargument to AI job-destruction predictions — if AI makes legal or financial analysis cheaper, more organizations will buy more of it.

A June 29 Fortune analysis confirms the Jevons effect probably holds for aggregate professional demand. The critical qualifier is who captures the gains:

• Entry-level professional services hiring has dropped 29% since January 2024 — what analysts are calling the worst entry-level market in 37 years.
• Workers aged 22–25 in AI-exposed occupations have seen a 13% employment decline since 2022 — consistent with the Stanford finding reported earlier in this post.
• Entry-level roles are now 7× more likely to require skills that historically appear later in a career — strategic decision-making, stakeholder management, judgment — compared to pre-AI job descriptions.
• Wolters Kluwer found AI produces professional-quality output on individual tasks 50–60% of the time, but only around 2% success when executing complete, end-to-end projects.

The picture this paints: the Jevons effect may be real for the profession over the next decade — more total legal work, more financial analysis, as AI lowers the cost floor. But those benefits accrue to a smaller, more senior workforce. A world with more lawyers but fewer law firm associates is a Jevons win for the profession and a structural loss for the first rung of the career ladder. The earlier advice in this post applies with more urgency: get visible at the AI-augmented level now, because the entry door is narrowing, not opening.

Ford's $4.8 billion lesson: AI needs human expertise to function

The most concrete real-world illustration of AI's current limits comes from Ford, which spent three years quietly rehiring 350 veteran engineers — internally called "gray beards" — after AI-driven quality failures became severe enough to threaten the company's competitiveness.

By mid-2024, Ford's quality issues were costing $4.8 billion annually in recalls. By July 2025, the company set a record with 90 recalls in a single year, including a $570 million charge on nearly 700,000 crossover vehicles. The root cause: AI systems trained on incomplete data couldn't identify failure points before manufacturing. Ford's VP of Vehicle Hardware Engineering Charles Poon identified the lesson: "Artificial intelligence is a fantastic tool, but it's only as good as the information you use to train it."

After bringing back the experienced engineers, Ford jumped from 10th to 1st place in JD Power's 2026 Initial Quality Survey.

Ford's experience fits a broader pattern. MIT's 2025 "GenAI Divide" study — which examined 300 corporate AI initiatives and interviewed 150 executives — found 95% of organizations saw zero meaningful return on their generative AI pilots. The failure was almost never the model. It was data readiness, workflow integration, and the absence of defined outcomes before build-start. The 5% that succeeded were those that preserved and leveraged human domain expertise — not those that tried to replace it.

For knowledge workers watching this from the outside: being the person whose expertise trains, audits, and guides AI systems is not a transitional role on the way to obsolescence. Ford's "gray beards" aren't a rearguard action — they're the reason the AI works at all.

July 2 update: The Remote Labor Index puts a number on it

The June 29 section above cited a Wolters Kluwer finding: AI produces quality output on individual tasks 50–60% of the time — but only around 2% when executing complete, end-to-end projects. As of this week, that 2% has a direct update from a benchmark built to measure exactly that.

The Remote Labor Index (RLI) — developed by the Center for AI Safety (CAIS) and Scale Labs (Scale AI) — tests whether AI agents can complete entire real-world freelance projects at a quality a paying client would accept. Human professionals evaluate every deliverable against a gold-standard reference produced by a paid expert. The benchmark covers 240 real projects across eight domains: 3D and CAD, architecture, graphic design, video and animation, audio, data analysis, web apps, and related work — representing over $143,000 in freelance earnings at market rates.

Results as of July 1, 2026:

Model	End-to-end automation rate
Claude Fable 5	16.1%
Claude Opus 4.8	8.3%
GPT-5.5	6.3%

When the benchmark launched in October 2025, the best-performing model completed just 2.5% of projects. In under eight months, the frontier rate has risen roughly 6×.

What 16.1% means — and what it doesn't

What it means: AI now completes roughly 1 in 6 real freelance projects at professional standard. Eight months ago, that was 1 in 40. The pace of change is as significant as the number itself — the benchmark authors call this "a significant increase in digital labor automation."

What it doesn't mean: The same authors are explicit about the ceiling: "Today's AIs still fall short of professional quality on most projects; none of the three Fable 5 deliverables above would be accepted as finished work." Eighty-four percent of projects still fail the professional bar. The hardest tasks involve multi-step editing, precise adherence to a detailed brief, and natural use of professional software — the skills that take years to develop and that AI agents remain weakest at.

One measurement nuance: only 218 of 240 projects were evaluated for Fable 5 before its government-access suspension. Even if Fable 5 had failed all 22 unevaluated projects, its automation rate would be 14.6% — still the highest on record.

What this adds to the overall picture

The RLI answers a question the other data in this post doesn't directly reach: not "what do employment statistics show" or "what do Americans fear" — but "how much actual professional output can AI produce on real tasks right now?" The answer as of July 2026: about 1 in 6 complete freelance projects, up from 1 in 40 nine months ago.

For freelancers and professionals doing project-based work: the 84% where AI still falls short tends to be work requiring domain expertise, multi-step judgment, and professional-tool fluency — the kind that takes years to develop. The 16% AI can now complete at professional standard leans toward simpler briefs and creation-from-scratch tasks. Knowing which category your work belongs to is an increasingly important question.

The pace matters as much as the number. If the October 2025–July 2026 trajectory continues — and it may not, since improvements in hard agentic tasks often plateau — AI doing professional freelance work at meaningful scale shifts from theoretical to a near-term question, not a decade-away one.

July 12 update: The AI-jobs pivot — and what it actually tells us

Two stories published today add new texture to everything above.

Sam Altman now says AI has been "net job-creating"

In a post on X, OpenAI's CEO wrote: "so far at least, i'm pretty sure AI has been net job-creating. this was not what i expected — although i was much less pessimistic than others, i thought by this level of capability we'd have seen some impact. it is possible this direction keeps going!"

This follows remarks Altman made at a Commonwealth Bank of Australia conference in Sydney on May 26, 2026, where he told CBA's chief executive Matt Comyn that OpenAI was "pretty wrong" about AI's economic impact, while "roughly right" on the technology itself. On his earlier prediction that entry-level white-collar jobs would be quickly eliminated: "I'm delighted to be wrong about this."

What "net job-creating" actually means. The aggregate framing is fair — and simultaneously misleading for workers most affected. Goldman Sachs's AI jobs tracker gives the cleaner breakdown: AI is currently eliminating roughly 11,000 US jobs per month in attributable layoffs, while data-center construction has added 212,000 positions since 2022 — about 9,000 per month — producing a slightly net-positive overall. The friction: those construction jobs don't replace the entry-level knowledge-work roles disappearing in marketing, customer service, design, and software. Goldman's own tracker finds a "slight positive correlation" between AI adoption rates and unemployment among workers under 30 — consistent with the Stanford and MIT entry-level data throughout this post.

The "net positive aggregate" and "concentrated entry-level pressure" can both be true, as they have been throughout this post. Altman's update tells us AI hasn't caused the broad unemployment crisis he once expected. It doesn't tell us the entry-level pressure has eased.

Experienced workers are choosing exit over adaptation

A separate Fortune story published today profiles experienced tech workers opting for early retirement rather than adapting to AI-driven workplace changes.

The reasons vary. Some, like software architect Jennifer Kerns, simply don't believe in AI: "That was really it for me. I don't buy into AI. I think it's a bubble that's going to burst." Others made an actuarial calculation: reskilling at 60-plus for a 3–5 year career runway often doesn't pencil out. An Allianz Life study (May 2026) found 42% of Americans retire earlier than intended — health, caregiving, and job loss remain the top reasons, but workplace technology disruption has become a new accelerant. Microsoft was separately reported to have extended buyout offers to approximately 7% of its US workforce in mid-2026, a move that skews heavily toward more experienced employees.

The workforce consequence is the one this post's Ford case study illustrates: when experienced workers exit, they take institutional knowledge with them. Researcher Chip McConnell framed it directly: "We are at risk of losing senior judgment necessary for ensuring AI matures healthily." For younger workers entering a market where senior mentors are departing at the same moment AI absorbs entry-level tasks, this is a double squeeze — less foundational work available to learn from, fewer experienced people to learn from while doing it.

What it actually means for you

• Don't panic from headlines. The aggregate data does not support "AI is taking everyone's jobs." Decisions made from fear tend to be bad ones.
• If you're early-career, treat the tools as urgent. The pressure is real at the entry rung, and the most reliable hedge is being visibly good with AI in your field — not avoiding it. Get hands-on fast.
• If you're established, move up the value chain. The work that's getting automated is the routine first-pass stuff. Being the person who reviews, directs, and judges AI output is more durable than being the person who produced the first draft.
• If you've already been hit, that's a different, tactical problem — see our 30-day AI-layoff reskilling playbook, and the agentic-AI job guide for the new roles that didn't exist three years ago.

The realistic posture isn't "AI will take my job" or "AI changes nothing." It's: the floor is shifting under early-career work specifically, the aggregate is stable for now, and the people who do best are the ones who get fluent with the tools instead of waiting to find out. New to those tools? Start with the AI Basics hub.

Sources

• Fortune: AI was supposed to be killing jobs. In spring, the labor market is opening up instead
• MIT Technology Review: A reality check on the AI jobs hysteria
• MIT Technology Review: It's time to address the looming crisis in entry-level work
• U.S. Bureau of Labor Statistics: Job Openings and Labor Turnover Survey (JOLTS)
• ADP: National Employment Report
• Semafor: Dan Schulman's plan to get Verizon back in the fight
• Hollywood Reporter: Verizon Cuts 13,000 Jobs, Sets Up $20 Million Reskilling Fund for AI
• TheStreet: Verizon CEO sends blunt warning on future of jobs after layoffs
• Inc.: Verizon CEO: AI Is Coming for Your Job, 'and Everyone Knows It'
• PYMNTS: Verizon's Dan Schulman Tells CEOs to Be Open About AI Job Cuts
• Bloomberg: Verizon CEO Sees AI Coming for Customer Service Jobs
• Ipsos: Reuters/Ipsos June 2026 Poll — 53% fear household job loss to AI, 71% fear permanent displacement, 73% concerned about AI use overall
• Yahoo Finance / Reuters: 53% of Americans fear AI could take their jobs, poll finds
• Decrypt: Americans Fear Job Losses Due to AI But Hope for Cancer, Alzheimer's Cures: Anthropic Survey
• The Decoder: Over half of Americans fear losing both their jobs and their independent thinking to AI, survey finds
• Fortune (June 14, 2026): AI job disruption is here. The problem may be compounded because nearly 75% of people don't apply for unemployment benefits
• allwork.space: U.S. Jobless Claims Remain Below Stress Levels Despite AI Layoffs
• Fortune (April 28, 2026): 'The cost of compute is far beyond the costs of the employee': Nvidia executive says right now AI is more expensive than paying human workers
• Tom's Hardware: Nvidia exec says AI is more expensive than actual workers — yet some companies don't see the extra costs as a negative
• U.S. Bureau of Labor Statistics: 2023 Survey of Unemployment Insurance Claimants
• Fortune (June 29, 2026): The most reassuring argument about AI and jobs quietly explains why Gen Z can't get one
• Fortune (June 29, 2026): Ford realized AI wasn't capable of taking human jobs years ago—and hired 350 'gray beard' engineers to steer its program
• Fortune (April 28, 2026): A 160-year-old paradox explains why AI will create more jobs, not fewer, top economist says
• Fortune / MIT Project NANDA (August 2025): 95% of corporate generative AI pilots fail to deliver ROI, MIT study finds
• Center for AI Safety (CAIS) and Scale Labs (July 1, 2026): A Significant Increase in Digital Labor Automation — RLI July 2026 results showing Fable 5 at 16.1%
• Scale AI: The Remote Labor Index: Measuring the Automation of Work — full benchmark methodology and original October 2025 results
• Scale Labs: Remote Labor Index Leaderboard
• ZDNet (July 2, 2026): Fable 5 just set a new AI freelance work performance record — but it can't replace humans yet
• Sam Altman on X (July 12, 2026): "so far at least, i'm pretty sure AI has been net job-creating"
• Euronews (May 26, 2026): No AI 'jobs apocalypse' so far, says OpenAI's Sam Altman
• The Decoder (July 12, 2026): OpenAI CEO Altman is now "pretty sure" AI is net job-creating
• Yahoo Finance (July 2026): Goldman Sachs AI Jobs Tracker — Altman says OpenAI was "pretty wrong" on job losses
• Fortune (July 12, 2026): More tech workers are retiring early because they don't want to deal with AI-related changes