AI Is Replacing Jobs. Why Aren't Defaults Rising?

The Gap Between the Headlines and the Data

Everyone is talking about AI replacing jobs. METR task horizons are doubling every 4 months. Challenger reports roughly 4,900 AI-attributed layoffs in 2025. The capability curve is moving faster than at any point in recent memory.

The household financial distress data hasn’t flinched.

The American Distress Index reads 64.0, Elevated. Driven by interest rates, depleted savings, and tightening credit. AI isn’t in the number yet. The ADI tracks Federal Reserve data with a 4-5 month lag. A worker who loses their job to AI today won’t appear in delinquency data until next quarter, and won’t show up in default statistics until the quarter after that. For the deeper mechanics of this pipeline, see When AI Takes the Job, Who Misses the Mortgage Payment?.

The Upstream Signals

AI Capability Growth

METR time horizons, a measure of how long tasks AI models can autonomously handle, are doubling every 4 months. The latest frontier model (Claude Opus 4.6) can handle tasks lasting 14 hours 30 minutes, up from 3.5 minutes in early 2023. METR’s own task suite is nearly saturated. The confidence interval on their latest measurement stretches from 6 to 98 hours.

When the autonomous task horizon was 3 minutes, AI could replace data entry and simple classification. At 6 hours, it replaced junior analyst work. Code review, document drafting. At 14+ hours, it can handle full-day professional tasks autonomously: multi-step research workflows, end-to-end software features, complex document production. These are not entry-level jobs. These are mid-career knowledge worker tasks.

The 80% reliability horizon, the task length where models succeed 80% of the time, also moved, from roughly 30 minutes to over an hour. Reliability, not peak capability, is what drives enterprise substitution decisions.

AI-Attributed Layoffs

Challenger, Gray & Christmas tracks announced layoffs and, starting in 2023, began categorizing those attributed to AI. In 2025, roughly 4,900 layoffs were explicitly AI-attributed. About 4.5% of total reported cuts.

That’s a small fraction. But it’s not zero. And it’s growing. In May 2023, when ChatGPT was 6 months old, IBM announced a hiring freeze affecting 7,800 positions that “could be replaced by AI.” Dropbox cut 500 jobs in 2023, citing AI efficiency gains. Duolingo replaced contractors with GPT-4 for content generation.

Tech Sector Contraction

BLS JOLTS data shows Information sector job openings (NAICS 51) collapsed from 218,000 in June 2024 to 122,000 in November 2024. A 44% drop in 5 months. AI is one factor among several. Post-ZIRP correction, over-hiring pullback, and efficiency drives all contribute. But the timing overlaps with rapid AI deployment.

The Downstream Data

Youth Unemployment

BLS unemployment for ages 20-24 sits at 7.1%, compared to 3.7% for ages 25-34. Young workers, who disproportionately work entry-level jobs most exposed to AI replacement, face double the unemployment rate of workers 5 years older.

What the Numbers Say Now

The data doesn’t show a clear AI-to-default connection yet. AI-attributed layoffs are 4.5% of total reported layoffs. That’s not enough volume to move the national ADI composite.

But the pipeline is forming. METR time horizons at 14 hours 30 minutes, doubling every 4 months. Roughly 4,900 AI-attributed cuts in 2025. ADI at 64.0, elevated but not spiking. Each stage of the pipeline is measurable. None have reached the threshold where they move national distress data.

The 3-6 month lag between job loss and default means if AI displacement accelerates in Q1 2026, it would show up in ADI by Q3 2026. Workers who face income disruption can find actionable steps in our guide on what to do when you’re behind on your mortgage.

The American Worker Index (AWI) now tracks this pipeline explicitly: AI capability growth → job displacement → youth unemployment → household financial stress. The AWI currently reads 83.0, in the Crisis zone.

The gap between AWI 83.0 and ADI 64.0 is the lag between the displacement and its downstream consequences. That gap is either going to close, meaning the distress pipeline activates at scale, or the economy absorbs the displacement faster than the capability curve accelerates it. Both outcomes are measurable. Both will show up in the data before they show up in headlines. For current labor market data, see our Unemployment & Job Market Statistics roundup.