Ask whether AI is taking jobs and you will get answers ranging from calm reassurance to genuine alarm, often from sources that both look credible. The honest answer requires holding two real, well-sourced, and seemingly contradictory findings at once.
The alarming data point
Challenger, Gray & Christmas, the standard source for tracking employer-stated reasons for layoffs, found AI was cited in 87,714 job cuts in the first half of 2026 alone, already surpassing all of 2025's total. In May 2026, AI became the single most-cited reason for layoffs that month, accounting for 38,579 cuts, the highest monthly total since the firm began tracking AI as a specific category in 2023. It has now been the leading cited reason for four consecutive months.
The calming data point, from an equally credible source
Yale's Budget Lab, using actual government labor statistics rather than employer-stated reasons, reached a strikingly different conclusion. Their analysis found that occupations with the highest AI exposure show no disproportionate job loss or unemployment spike compared to low-exposure occupations, roughly 33 months after ChatGPT's launch. The researchers are careful to note this is a snapshot, not a prediction that disruption will not come later, but as of their most recent data, the aggregate numbers simply do not show it yet.
How both of these can be true at once
The gap is explained by what each source is actually measuring. Challenger tracks what companies say publicly, and companies do not always tell the truth about their own layoffs. When Amazon cut roughly 30,000 corporate jobs across late 2025, CEO Andy Jassy had earlier told employees in an internal memo that AI would shrink the workforce, but when discussing the actual cuts, he said they "wasn't triggered by financial strain or artificial intelligence," directly contradicting his own earlier framing. IBM faced a viral claim that it laid off 8,000 HR workers and replaced them with an AI chatbot; CEO Arvind Krishna told the Wall Street Journal the real number was in the hundreds, not thousands, and that IBM's total headcount actually grew during that period, with savings redirected into engineering and AI development roles. A January 2026 Harvard Business Review piece summarized the pattern directly: "companies are laying off workers because of AI's potential, not its performance."
The group actually showing measurable impact
This is the finding most coverage misses entirely. Stanford's Digital Economy Lab analyzed millions of real ADP payroll records and found a specific, age-skewed effect: employment for workers aged 22 to 25 in AI-exposed occupations, particularly software engineering and customer service, has declined 13% relative to older workers in the same roles since late 2022, and the decline is accelerating, from 2.8% a year to 3.8% a year in the most recent data. This reconciles the two seemingly contradictory findings above: the disruption may be real but concentrated narrowly in entry-level, AI-exposed roles, small enough to stay invisible in aggregate national statistics while being very real for the specific group experiencing it.
What real usage data actually shows
Anthropic's own Economic Index, based on real, anonymized Claude conversation data rather than a survey, found that after a brief period where fully automated tasks outpaced human-assisted ones, the balance shifted back by late 2025 to 52% augmented and only 45% automated, meaning more than half of real AI use is assisting a human with a task, not replacing them outright. The same data shows usage diversifying beyond its early concentration in coding, with the top 10 most common task types falling from 24% to 19% of total usage between November 2025 and February 2026.
What history actually suggests, with an honest caveat
The most commonly cited historical parallel is bank tellers and ATMs. ATMs reduced tellers per branch from 21 to about 13, but made branches cheap enough to open that banks opened far more of them, urban branch counts grew 43%, and total teller employment ultimately grew, with the job itself shifting from cash handling to advisory work. But this comparison has a real limit worth stating honestly: prior automation waves hit less-educated, manual-labor jobs hardest. AI exposure research from the International Labour Organization finds the opposite pattern this time, exposure skews toward more educated, white-collar work, meaning AI may not distribute its disruption the same way past technology did, even if the aggregate, long-run outcome eventually looks similar.
The takeaway
The honest answer to "is my job safe" is not yes or no, it is that the disruption so far is real but narrower and more age-concentrated than the layoff headlines suggest, companies are not always accurate about why they are actually cutting jobs, and the group with the clearest, most measurable impact right now is workers at the very start of their careers in AI-exposed fields. That is a more useful thing to know than either a reassuring or an alarming headline number.
