AI Layoff Tracker

What the summary cards mean. Verified job cuts is the main figure: cuts with a filing or independently reported source behind them. Explicitly AI-attributed is a subset of Verified job cuts where the source explicitly names AI as a cause. Announced job cuts is a separate announcement-history figure: source-linked plans reported at announcement stage. A later filing or report is linked or merged when confidently matched; an unmatched announcement is not a claim that cuts remain unexecuted. Announced cuts are not counted in Verified or AI-attributed totals, so the cards do not double-count.

Geography in the cards. Country and US-state filters describe the documented location of affected jobs, not an employer’s headquarters or every place it operates. A national announcement without a source-supported job-location state remains state-unspecified rather than being assigned to a state by inference.

What this is. A continuously updated, source-linked database of publicly reported layoffs worldwide. It records the source, evidence quote, event status and revision history so every figure can be independently checked. It is not a claim of complete coverage in every country.

Where the data comes from. Sources are always labeled on the entry: SEC filing legal 8-K and 6-K filings pulled from SEC EDGAR full-text search (strongest evidence; US public companies and foreign private issuers that file with the SEC). WARN notice state government mass-layoff filings from 42 covered US states. Company statement reviewed investor-relations and newsroom feeds. News named reports discovered through GDELT and NewsAPI, then retained only when the record has usable evidence. Eurofound ERM is a separately labeled, thresholded European announcement source.

How often it updates. News and SEC filings: twice daily (morning and after US market close, ET). WARN notices: daily at 11 AM ET, sweeping every covered state. An automated anomaly review runs daily at noon ET, flagging statistically unusual entries (very large single notices, same company filing in several states, weak source links) for human inspection before anyone else finds them.

How entries are extracted and checked. News and filings are machine-extracted; core facts must appear in the source text. Counts parse conservatively (ranges resolve to the lower bound). Countries and industries normalize through fixed vocabularies; implausible values are rejected. New records carry an evidence confidence and publication status. Exact fingerprints, same-company guards and cross-source comparison prevent double counting; uncertain candidates remain provisional instead of silently inflating verified totals. WARN filings skip the language model and remain exempt from fuzzy dedup because one employer can legally file several distinct notices.

How the AI tag works. We distinguish AI as a primary cause, a contributing cause, a selection/operations tool, background context, and an explicit denial. Only primary or contributing cause classifications may be AI-attributed, and each must carry an exact supporting quote found in the source text. AI investment, future automation projections, and AI used to select workers do not qualify by themselves.

Coverage and honest limitations. US depth is greatest because of WARN and SEC sources. Europe has structured coverage of large announcements through Eurofound ERM. Outside those live collectors, country-level coverage is currently worldwide news discovery and any explicitly reviewed company newsroom feed; named filing systems such as SEDAR+, RNS, ASX, TDnet and HKEXnews are research candidates, not silently assumed feeds. WARN and ERM have their own thresholds and geography rules, so they should not be summed as if they were a complete national census. Multi-state and multi-country events can overlap; the entry and source fields disclose that risk. Entries dated in the future are announced or filed but not yet completed.

What we exclude. Rumored or unsourced layoffs; layoffs with no stated job count; forward-looking projections (e.g. “could cost X jobs by 2050”) rather than announced or executed cuts; and retrospective summary articles that would double-count events already tracked.

Why our totals differ from other headline numbers. Three kinds of trackers measure three different things. Government statistics (BLS) count every separation in the economy, millions per month, with no event-level detail. Announcement surveys (Challenger, Gray & Christmas; the WSJ and TrueUp trackers) count corporate intentions: when a CEO announces “20,000 cuts over the next two years,” the full 20,000 lands in their total that day, even though much of it may come through attrition, get scaled back, or never produce a single filing. This tracker counts only what has a verifiable document or quoted primary source behind it: the WARN notices and SEC filings that appear as those 20,000 cuts actually execute, plus reported cuts with a named-outlet source. A worked example: in the first half of 2026, announcement surveys reported roughly 443,600 US job cuts (Challenger, through June); verified filings and sourced reports here totaled about 175,000 for the same period, both correct answers to different questions. Theirs answers “what are companies saying?” Ours answers “what can you prove?” Treat our verified figure as a documented floor: smaller than the estimates, but every single number is clickable back to a legal filing or named outlet. Since July 2026 we also track announcement-stage cuts as their own labeled tier (“Announced”, tagged in the table and shown as a separate headline number) so both questions are answered on one page, and unlike the announcement surveys, every announcement here links to its source too.

Using the data. Free with attribution to asktherecruiter.com. The CSV and JSON buttons download exactly what your current filters show (or the full dataset when unfiltered); each chart offers its own image or data download. Programmatic access: GET /blog/wp-json/layoffs/v1/query (paginated; filter params match the page: years, quarters, months, industry, country, state, sources, reasons, q, from, to) and GET /blog/wp-json/layoffs/v1/aggregate for totals and breakdowns. Corrections get priority via the contact page or [email protected], and every fix is disclosed in the corrections log below.

Official government filings and notices are collected directly (SEC EDGAR incl. Item 2.05 exit-cost filings, WARN notices from 44 US jurisdictions, Eurofound ERM for the EU, discovery probes for Japan, South Korea and Brazil), press-release wires and reviewed company IR feeds are monitored, and 499 reviewed news outlets across 198 countries surface coverage through GDELT’s 65-language index and NewsAPI — allowlist-only, never crawled directly. Every published event links to its source.

Checking the most recent collector status…

“Healthy” means the collector completed and reports how many candidate documents it found; it does not mean the source is a complete census. “Running” means collection is in progress. “Degraded” means the most recent attempt failed, so that source should not be interpreted as reporting zero layoffs. The full corrections trail is in the Data notes & corrections log below.

Why the figures differ. The cards above are scoped by the job-location country filter, while Challenger measures announcements by US-based employers. They are therefore not a like-for-like Challenger total. This is a transparent coverage comparison, not an accuracy score and not a command to change our totals. Two labeled pairs are compared, each updated automatically when Challenger publishes its monthly report: Challenger AI cuts vs AskTheRecruiter announced AI cuts (strict), and Challenger all announced cuts vs AskTheRecruiter announced US cuts. The strict AskTheRecruiter figures include only canonical events with a source-evidenced announcement date, a US-based employer and announcement-stage status (plus AI as the primary stated cause for the AI pair). The wider job-location/any-AI figure is diagnostic only and is not comparable to Challenger.

The teal “AI-linked, broad (Challenger-style)” line measures AI the way Challenger and layoffs.fyi do: it adds events where the company or press tied the cuts to AI loosely, including layoffs made while funding an AI pivot and AI-driven market disruption. Our strict AI tag (the employer’s own words, quote on file) stays separate and unchanged; the broad line exists so the two counting philosophies can be compared side by side on the same chart.

Monthly, source-linked announcement figures. This is a coverage reconciliation, not an accuracy score.

Cumulative year-to-date values use the same reference months and official reports.

The current gap principally reflects incomplete source-evidenced announcement-date and employer-domicile enrichment, plus different source mixes. It does not mean that untracked events are false, that tracked events are wrong, or that the tracker is complete. Each monthly report remains available here even when the workflow flags a large gap for investigation.

The region tabs are views over the worldwide data. The full country list for each tab loads here.

Different trackers measure different things. Here are the factual differences, so you can pick the right number for your purpose.

Challenger, Gray & Christmas reports monthly totals of announced US job cuts, compiled from press reports and company statements, including estimates and multi-year plans. It is published as monthly press releases with no per-event public database. Because announcements exceed executed cuts, Challenger’s totals run above any verified-event count, including ours.

WSJ layoffs coverage is editorially selected major announcements with newsroom verification. There is no downloadable dataset, and coverage is selective by design.

TrueUp and Layoffs.fyi are technology-sector trackers built from announcements and crowdsourced reports, and they are downloadable. Their scope corresponds to our Technology industry filter, not our all-industry total.

Official statistics such as US BLS JOLTS (all separations economy-wide, millions per month), UK ONS redundancies, and Eurostat are survey-based aggregates with no company-level detail. Event trackers measure a different universe and will not match them.

This tracker puts verified events only in the headline totals (filings and sourced reports, each one linked), keeps announcement-stage figures in a separately labeled tier, discloses corrections automatically below, and publishes its data and code. When our number differs from a tracker above, the difference is definitional, and both definitions are stated here so either number can be used correctly.

United States and SEC-reporting foreign issuers. SEC EDGAR full-text search for 8-K and 6-K filings, searched twice daily across layoff phrasings, plus official state WARN notices from 41 states, imported daily. The complete per-state list of portals and parsers is public in our source code. The remaining states publish no usable per-notice data: HI and OK omit headcounts, and MO and NM publish nothing.

European Union, Norway, and the UK historically. The European Restructuring Monitor from Eurofound, an EU agency. These are per-company restructuring announcements compiled by national correspondents who screen 58 designated business media titles daily. We import them daily with attribution. Because ERM records announcement-stage figures, its entries feed our Announced tier.

Worldwide, every country. The GDELT global news index machine-translates press coverage from 65+ languages (Le Monde, Handelsblatt, Nikkei, Globo and thousands more) and we search it twice daily for layoff coverage in any country, alongside NewsAPI. Only articles from an editorially maintained trusted-outlet list are ingested, and that list is also public in the repository.

What the AI does and does not do. WARN and ERM records are imported with no AI processing, because they are already structured. For press articles, the DeepSeek-V3 model reads the article text and extracts the company, the count, the date, the country, and any explicit AI attribution. Countries and industries then normalize through fixed vocabularies, counts and dates pass hard validation rules, and duplicates are checked against the existing data. A second, independent model pass audits classifications every day, and a full-dataset audit runs monthly. Label corrections apply only when two independent passes agree. Numeric changes and removals always require a human. Every correction discloses itself in the log below.

Why 41 US states and not 50. All 50 states appear in the tracker through SEC filings and news, but WARN notices come from 41 states because that is how many publish usable per-notice data. The rest cannot be included through no fault of ours: Hawaii and Oklahoma publish WARN notices without headcounts, and Missouri and New Mexico publish nothing citable at the notice level. 41 is the ceiling of what US states actually make public, and we are at it.

Why the country count grows over time. The number of countries is not a setting we can raise; it reflects where large, press-covered layoffs have actually happened in our window. GDELT already searches every country on earth in 65+ languages, so a country appears the moment a credible outlet there covers a qualifying layoff. As events occur and as we add more trusted local outlets (the list grew from 106 to 153 in July 2026), the count rises on its own. This is honest by design: we show the countries where verifiable events exist, not a padded list.

Known gaps, stated plainly. We do not yet operate direct connectors for Canada SEDAR+, UK RNS, ASX, TDnet/EDINET, NSE/BSE, HKEXnews, SGXNet, SENS, DART or TASE; they are maintained as official-source research candidates and will be named as live only after a stable public interface, tests and source-health monitoring exist. A few countries also publish official per-company redundancy records we do not ingest yet, including Belgium’s FPS Employment collective-dismissal reports, Italy’s weekly CIGS decree lists, and Sweden’s varsel statistics. Most countries, including Germany and Mexico, treat employer identity in redundancy filings as confidential, so press coverage through GDELT in local languages is the primary source there. Events too small for any press coverage, any WARN threshold, or the ERM threshold of 100 jobs will not appear in any tracker, including this one.

Errors are corrected openly, not silently. Every correction to published figures is dated and described here, newest first, and corrected rows are also flagged edited: true in the API. The list scrolls, because it grows a little every day as the data self-corrects.

For reproducible monitoring, the machine-readable quality status endpoint reports dataset revision, recent corrections, collector health, retained-source integrity and the status of each coverage workstream. Pending work is shown as pending—not silently treated as coverage.

• 2026-07-19: 16 entries reclassified. Automated source-evidence reassessment
• 2026-07-19: 2 entries enriched. Automated source-evidence context enrichment
• 2026-07-19: 2 entries reclassified. Automated source-evidence reassessment
• 2026-07-19: 4 entries reclassified. Automated source-evidence reassessment
• 2026-07-19: 2 entries reclassified. Automated source-evidence reassessment
• 2026-07-19: 7 entries reclassified. Automated source-evidence reassessment
• 2026-07-18: 1 entry enriched. Automated source-evidence context enrichment
• 2026-07-18: 1 entry enriched. Automated source-evidence context enrichment
• 2026-07-18: 1 entry corrected (fields: job_count, excerpt). R8 Oracle reconciliation: converge on the company-grounded figure. Oracle's FY2026 10-K discloses headcount fell from ~162,000 (May 31, 2025) to ~141,000 (May 31, 2026), a net ~21,000 reduction the filing links to 'the adoption and deployment of AI technologies across our operations'; the prior 30,000 was press-reported gross plan size originating from a TD Cowen analyst estimate. Gross plan repor
• 2026-07-18: 1 entry merged. R8 Oracle reconciliation: the Feb 20,000 row is analyst-forecast coverage (TD Cowen 20K-30K, 'may lay off up to 30,000') of the same Fiscal 2026 Oracle Restructuring Plan executed Mar 31, 2026 and held as the Apr 6 30,000 row; a single plan was double-counted as 50,000. Both Times of India reports are retained on the canonical event.
• 2026-07-18: 1 entry corrected (fields: job_count, country, excerpt). Dow Jan-2026 correction (plan R2 #3, amended): count restored to 3,700 – ERM factsheet 204559 books the net-new portion of the 4,500-job global plan announced Jan 29, 2026 (CBS); the remaining 800 were announced July 2025 and separately retained (row 61702 / ERM 204001), so 4,500 double-counts. The CBS source-attach step had overwritten the unpinned row to 4,500 and blanked country; both restored
• 2026-07-18: 1 entry reclassified. Automated source-evidence reassessment
• 2026-07-18: 1 entry enriched. Automated source-evidence context enrichment
• 2026-07-18: 1 entry enriched. Automated source-evidence context enrichment
• 2026-07-18: 1 entry reclassified. Automated source-evidence reassessment
• 2026-07-18: 1 entry corrected (fields: job_count, layoff_date, announced, ai_explicit, ai_causation, review_status, excerpt). R2 #1 Intuit correction: extractor parsed '17% of workforce' as 17 jobs. Corrected to the article-stated ~3,000 ('Intuit is letting 17% of its staff go, or about 3,000 people' – TechCrunch 2026-05-20; 'approximately 3,000 employees' – CFO Dive). Original CNBC URLs are 403-blocked; count evidenced via accessible allowlisted alternates. AI fields set to the company's explicit on-record denial: CEO s
• 2026-07-18: 1 entry corrected (fields: job_count, source_url, excerpt). Meta 2026-03-25 Reality Labs/recruiting/sales round undercounted: extractor stored 100 from a 'hundreds' headline (economictimes). Corrected to 700 per The Information/Bloomberg-sourced reporting ('about 700'); primary CNBC 2026-03-25 article is 403-blocked, accessible alternate quoted. Challenger gap plan R2 #2. Original source report retained on event 43623.
• 2026-07-18: 1 entry merged. R2 #4 Lucid duplicate: 'Lucid' (livemint) and 'Lucid Motors' (TechCrunch) 1,500-job rows for the 2026-06-22 announcement are the same event (18% cost-reduction round, same day); company_key normalization keeps 'motors' so dedupe never clustered them. Merged into the TechCrunch-sourced canonical row retaining both source reports.
• 2026-07-18: 14 entries reclassified. Automated source-evidence reassessment
• 2026-07-18: 9 entries reclassified. Automated source-evidence reassessment
• 2026-07-18: 4 entries reclassified. Automated source-evidence reassessment
• 2026-07-18: 2 entries corrected (fields: industry, country). Automated classification audit: label mismatch flagged and independently confirmed by two LLM passes (DeepSeek)
• 2026-07-18: 28 entries merged. Daily cross-source dedup: same layoff event reported by multiple sources, confirmed by DeepSeek
• 2026-07-18: 1 entry reclassified. Automated source-evidence reassessment
• 2026-07-18: 1 entry removed. Removed structurally blank legacy row with no company, date, source name, or source URL; confirmed by public integrity sample 2026-07-18
• 2026-07-17: 2 entries enriched. Automated source-evidence context enrichment
• 2026-07-17: 4 entries corrected (fields: industry, country). Automated classification audit: label mismatch flagged and independently confirmed by two LLM passes (DeepSeek)
• 2026-07-17: 26 entries merged. Daily cross-source dedup: same layoff event reported by multiple sources, confirmed by DeepSeek
• 2026-07-17: 3 entries enriched. Automated source-evidence context enrichment
• 2026-07-17: 55 entries merged. Daily cross-source dedup: same layoff event reported by multiple sources, confirmed by DeepSeek
• 2026-07-16: 1 entry corrected (fields: country). Automated classification audit: label mismatch flagged and independently confirmed by two LLM passes (DeepSeek)
• 2026-07-16: 34 entries removed. Daily cross-source dedup: same layoff event reported by multiple sources, confirmed by DeepSeek
• 2026-07-16: 1 entry corrected (fields: industry). Automated classification audit: label mismatch flagged and independently confirmed by two LLM passes (DeepSeek)
• 2026-07-16: 50 entries removed. Daily cross-source dedup: same layoff event reported by multiple sources, confirmed by DeepSeek
• 2026-07-15: Florida test rows removed, 87,600 jobs. Florida’s official WARN export contains internal test entries, which are fictitious notices sharing one WARN number and using non-existent zip codes. Eight such rows were removed, the largest a fake 78,788-worker “AT&T” notice that briefly ranked as our biggest entry. Our importer now skips test-named rows, and each removed row is permanently blocked from re-import.
• 2026-07-15: Country assigned to 88 news and SEC entries. These rows had no country recorded, which hid them from the regional views and country charts, though they were always in the worldwide totals. Each was resolved from its own source article. The largest were Oracle (30,000, spanning the US, India, Canada, Mexico and Uruguay, so “Multiple countries”) and BBC (2,000, United Kingdom).
• 2026-07-15: Ideal US Talent Systems RI corrected from 9,891 to 2. The Rhode Island notice states the company-wide figure with only 2 RI employees affected, and the per-state filings for DC, GA, IL and VA are already separate entries. Counting the company-wide total under RI double-counted the event.
• 2026-07-15: Ten non-events removed. These were SEC-filing extraction mistakes: severance dollar figures and workforce-reduction percentages misread as headcounts, WARN Act boilerplate clauses from acquisition agreements, and three duplicate rows of one Meta story carrying wrong dates.

Spotted something off? Every entry links to its primary source so you can check us. Send corrections through the contact page or to [email protected], and they get priority.