30MAR

Columbia: 75% of jobless never file claims

2 min read

08:00UTC

The workers AI displaces are precisely the categories the unemployment system cannot see.

← Back to Trump wants Iran's oil; 3,500 Marines land Jump to analysis ↓

ConflictAssessed

Key takeaway

Three quarters of displaced workers never appear in the claims data policymakers depend on.

Fortune and Columbia University research confirmed that roughly 75% of unemployed Americans never file for unemployment insurance⁠¹. Severance packages delay filing. Recent graduates lack sufficient work history to qualify. Contractors are categorically ineligible.

Initial jobless claims fell to 202,000 for the week ending 28 March, a 10-month low⁠². New York's updated WARN Act, the world's first law requiring companies to disclose AI's role in mass layoffs, produced zero AI attributions from 162 companies covering 28,300 workers after nearly a year. The nine-senator coalition pushing for expanded BLS data collection is demanding better measurement from agencies whose existing tools are structurally blind to what they are meant to measure.

The National Bureau of Economic Research survey confirmed another dimension: executives use AI only 1.5 hours per week on average, yet project a 0.7% employment decline over three years. Workers at the same firms expect a 0.5% increase. Those planning the cuts and those absorbing them hold irreconcilable forecasts.

Deep Analysis

In plain English

In the US, when you lose your job you can apply for unemployment benefits. Those application numbers are published weekly and are one of the most widely watched economic indicators. Politicians and journalists use them to judge whether the job market is in trouble. The problem is that about 75% of people who are unemployed never apply. If you get a severance package, you often wait until it runs out. If you are a freelancer or contractor, you are not eligible. If you just graduated and have not worked long enough, you cannot apply. These are exactly the kinds of people most likely to be affected by AI Job Displacement: well-paid tech workers with severance, contractors, and new graduates. They are invisible to the data that policymakers rely on.

Deep Analysis

Root Causes

The US unemployment insurance system was designed in 1935 for industrial workers in continuous employment. It requires sufficient recent earnings, excludes the self-employed and contractors, and has not been updated to reflect the rise of gig work, project-based employment, or high-earning knowledge workers who receive severance packages on exit.

The specific worker profile that AI is displacing in 2026 fits the non-filing demographic almost precisely. Senior software engineers at Oracle and Salesforce with severance packages will wait months before filing. Recent graduates without two years of employment history cannot file.

Contractors and freelancers are categorically ineligible. The system's design was never a problem when the workers it could not see were a minority. It becomes a structural failure when the invisible cohort is the one at the centre of the displacement wave.

What could happen next?

Policymakers relying on jobless claims as their primary AI displacement signal will persistently under-respond to the scale of the problem, delaying support programmes until the financial distress of displaced tech workers becomes visible through consumer credit defaults rather than benefit filings.

Source Landscape

This story draws on mixed-leaning sources from United States

United States×2

LeftRight

Consensus view: The Fortune and Columbia University research confirming 75% non-filing among unemployed Americans is the critical methodological problem in the current AI displacement debate. The three categories most likely to be displaced by AI (higher-earning tech workers with severance, contractors ineligible for unemployment insurance, and recent graduates with insufficient work history) are precisely the categories the claims system cannot see.

Counter-view: Initial jobless claims at 202,000, a 10-month low, are not purely a measurement artefact. They also reflect the genuine heterogeneity of the labour market: health care, construction, and logistics are hiring vigorously, and those workers, if displaced, would be more likely to file claims. The non-filing problem is real but concentrated in the specific worker profile AI is displacing, not universal.

Key tension: The nine senators led by Hawley and Warner are asking the BLS to collect better AI attribution data using a methodology built around claims and payroll surveys. But if 75% of AI-displaced workers never file, the underlying measurement tool is structurally unsuited to the task, regardless of how many new questions are added to it.

Sources:Department of Labor·Bloomberg·Fortune / Columbia University

Mentions:Fortune →Bureau of Labor Statistics →Senator Mark Warner →Columbia University →Department of Labour →National Bureau of Economic Research →New York →AI Job Displacement →

First Reported In

Update #4 · AI leads US layoffs as cuts go uncounted

Department of Labor· 4 Apr 2026

Read original →

Causes and effects

Caused by

NY AI layoff law: 162 filings, zero hits

New York WARN Act failure to capture AI attributions extends the measurement failure documented when 75% of unemployed never file.

Occurred 26 Mar 2026

Read story →

NBER: nine in ten firms untouched by AI

NBER finding that 90% of firms report zero AI impact persists because measurement systems cannot see the displaced workers.

Occurred 25 Mar 2026

Read story →

This Event

Columbia: 75% of jobless never file claims

Jobless claims data systematically undercounts AI displacement, undermining the metrics that policymakers rely on.

Led to

Stanford: AI costs 34 hires per layoff

The Fortune/Columbia finding that 75% of unemployed Americans never file for benefits established measurement blindness dimension one; JOLTS confirms hires-not-made are invisible by a further order of magnitude.

Occurred 10 Apr 2026

Read story →

Different Perspectives

Russia

Russia vetoed the same renewal on 17 September, arguing that Britain, France and Germany never validly triggered the snapback that reimposed the pre-2015 UN resolutions. No panel was ever seated under that mandate, so the UN list decays fastest for states that screen against it rather than against the American one.

China

China vetoed renewal of the UN sanctions monitoring mandate on 17 September, arguing that Resolution 2231 terminated on 18 October 2025 and that the Security Council should drop Iran's nuclear file altogether. On that reading there is nothing to monitor, so the sanctions survive and their enforcement does not.

Iraq

Baghdad saw the last American counter-Islamic State troops leave its territory on 30 September, completing a timetable it agreed with Washington in September 2024. Iraqi airspace deconfliction passes to Baghdad, which still has an open inquiry into the Maysan drone launches that has named nobody.

Pakistan

Treasury names Waseem Pasha Tajammal of Rawalpindi as the Cavalier group's chairman and places one of the designated incorporations in Islamabad. QatarEnergy separately told Pakistan that liquefied natural gas cargo cancellations would run through November, so Islamabad carries an enforcement question and a supply gap at once.

Turkey

Treasury named a Cavalier Dynamics company incorporated in Istanbul among the ten nodes it designated on 29 September, and Ankara has published no response. Turkey imported a record 120,000 barrels a day of Indian diesel in August, cutting Russia's share of its diesel imports to 20 per cent.

India

Suraj Yadav, a wiper from Uttar Pradesh, was killed aboard the Cape Dao on 23 September, and 19 of the ship's 20 Indian crew were taken off alive. India's September imports ran at 575,000 barrels a day from Iraq and 566,000 from Saudi Arabia, back to pre-conflict rates.