Newsom's 180-Day WARN Clock: The Last Clean Cal-WARN Sourcing Window

On May 21, 2026, one day after Meta began cutting 8,000 employees into Alexandr Wang's Superintelligence Labs, Governor Gavin Newsom signed Executive Order N-6-26. It does not change a single line of the Cal-WARN Act today. But it puts the richest pre-LinkedIn signal sourcers have on a 180-day clock toward an AI-specific overhaul, with recommendations due by mid-November 2026.

If you source displaced engineers off WARN filings, the next six months are the cleanest data you will ever get. After that, the cadence, the language, and the competition all change.

What the executive order actually does

EO N-6-26 directs two state agencies on two different clocks. The Labor and Workforce Development Agency (LWDA) has 180 days to review and recommend revisions to the Cal-WARN Act so it functions as an early-warning system for AI-driven displacement. Separately, the Employment Development Department (EDD) has 90 days to launch a public dashboard of AI's labor-market impact, drawn from Unemployment Insurance data.

Fisher Phillips and Sheppard Mullin both note what the order does not do: it creates no immediate legal obligation for private employers. SB 951, the pending statute that would create an AI-specific WARN-style notice, is the parallel legislative track. The EO is the agency homework assignment that feeds it.

For sourcers, the practical translation is simple. Today's Cal-WARN filings still flow under the rules you already know. Employers with 75 or more employees must give 60 days written notice of a mass layoff, plant closure, or relocation affecting 50 or more workers. SB 617 added a 2026 requirement that notices describe re-employment services and workforce board coordination. None of that changes for the next six months.

What changes is the field after mid-November.

Why the Meta filings are the cleanest example you'll get

Meta's May 20 announcement was the loudest AI-attributed cut of 2026. The cleanest piece of it, for a sourcer, is not the 8,000 number. It is the location-level Cal-WARN entries: 124 positions at the Burlingame office effective May 22, and 74 at Sunnyvale effective May 29.

Those two filings are the exact shape of signal that is about to get harder to read. Today they tell you precisely which campuses are bleeding, which gives you a tight geographic boolean and an obvious commute radius for outreach. After mid-November, if LWDA recommends AI-attribution disclosure and the legislature picks it up, expect employers to lawyer the language. Companies already avoid stating that an AI tool replaced a human. As Sheppard Mullin observed, executives "often use administrative phrasing." A "modernized" WARN with new disclosure fields will produce more text, not more clarity.

114,000

The Meta filings sit inside a much larger flow. Independent trackers logged more than 114,000 tech-sector losses across 150 companies in 2026, averaging about 825 per day. Microsoft launched its first-ever voluntary retirement program the same week as the EO, offering buyouts to roughly 7 percent of its US workforce. Oracle cut an estimated 30,000 in March. Amazon eliminated 16,000 corporate roles in Q1. AI is now the cited reason for more than a quarter of recent layoffs.

If your pipeline depends on reading WARN filings before LinkedIn updates, your edge is largest right now and shrinks every week toward November.

How sourcers actually use WARN today

The mechanics are well known to anyone who has done this work. WarnTracker.com, built by Steven Zhang and Chris Talley after Zhang's January 2023 Google layoff analysis crossed 600,000 LinkedIn views, aggregates publicly filed notices for free. WARN Firehose is the paid commercial layer, with SMS and email alerts, an API, and an SEC cross-reference at $499 a month on its Investor tier (the Pro plan is $49). Three states (Arkansas, Wyoming, New Hampshire) do not publish notices at all.

The standard playbook: filing drops, you pull the employer and the location, you build a boolean on LinkedIn restricted to that office's commute radius, and you reach out before the affected employees flip their headlines to "Open to Work." That window is usually 7 to 30 days.

Two things break this playbook when you scale it.

First, WARN gives you a company and a count, not a roster. You still have to guess which 124 of Meta Burlingame's roughly 2,000 employees are in the cut. Boolean queries on title and tenure get you a candidate pool, not a target list. This is where most sourcers burn hours, and where tools that operate in plain English start to matter. Describe what you want ("Meta Burlingame infra engineers, 3+ years, Python and Kubernetes, not yet marked Open to Work") and you should get a ranked shortlist, not 4,000 profiles to triage. Refolk was built for exactly that shape of query across GitHub, LinkedIn, and the open web.

Second, WARN systematically misses the most valuable engineers. The 75-employee and 50-layoff thresholds mean the Series B that quietly cuts a six-person AI team never files. Stanford HAI's 2026 AI Index reported that employment for software developers ages 22 to 25 has fallen nearly 20 percent from 2024, even as headcount for older developers in the same companies kept growing. A huge slice of that compression is happening below the WARN line, at startups and inside teams too small to trigger notice.

The 180-day playbook

Here is what to do with the window. None of it is novel, but the deadline forces priority.

1. Industrialize the alert layer now

If you are still checking WarnTracker manually, stop. Set up named-employer alerts on every Cal-WARN-filing company in your patch, plus the obvious AI-cited cutters: Meta, Microsoft, Oracle, Amazon, Cisco, Block, Intuit, Salesforce. WARN Firehose's API is the cheapest way to pipe filings into Slack or your ATS. The point is to react in hours, not on Monday-morning review.

2. Pre-build the rosters before filings drop

The filing tells you the office and the count. Your job is to already have a list of who works there. For every target employer in your patch, build and refresh a roster of engineers by team and location once a month. When the WARN hits, you are matching, not searching. This is the highest-leverage prep work you can do in the 180 days, because rosters built off today's clean data will still be useful in 2027.

3. Time outreach to the severance curve, not the filing date

Meta's package is 16 weeks of severance plus two more weeks per year of tenure. That means displaced engineers are not financially desperate for roughly four months. Recruiters who blast on day one get ignored. The reply rates climb in week 4 through week 10, when the severance shock has faded and the job search has gotten real but the runway is still visible. WARN gives you the date. Severance defines the urgency curve. Stagger your outreach accordingly.

4. Build the below-WARN pipeline in parallel

The startup cuts that never trigger filings are where the most interesting AI engineers actually live. You will not find them on WarnTracker. You find them by watching GitHub for commit gaps on previously active contributors, watching for sudden bio changes on personal sites, watching for company About pages quietly losing names. This is the work LinkedIn cannot do for you and dashboards definitely will not. It is also where plain-English search across GitHub and the open web (the second thing Refolk does well) outperforms boolean by a wide margin.

The dashboard will commoditize what every sourcer does manually. Differentiation moves upstream, to who is worth reaching and in what order.

5. Treat the EDD dashboard launch as a deadline, not a feature

Within 90 days of May 21, EDD must ship a public AI-jobs dashboard. The moment it ships, every recruiter, outplacement firm, journalist, and competitor sees the same view of which sectors and regions are bleeding. The advantage of mining WARN filings before LinkedIn updates collapses, because the dashboard is upstream of both. Your differentiation has to move from "I saw the filing first" to "I know which engineer on that team is worth reaching, with what message, in what order." That is a sourcing problem, not a data problem.

What "modernized WARN" probably looks like

Reading the EO text alongside SB 951, the likely shape of the November recommendations is some combination of:

• AI-attribution fields on the notice itself (was the role eliminated because of automation, augmentation, or unrelated factors).
• Lower thresholds for AI-driven cuts, possibly capturing smaller employers.
• Faster reporting cadence, possibly tied to the EDD dashboard feed.
• Mandatory coordination language with state retraining programs, extending the SB 617 disclosure logic.

Two of those help sourcers (lower thresholds, faster cadence). Two work against you (attribution fields that get lawyered into mush, retraining coordination that pulls candidates toward state programs and away from your reqs).

The net effect is probably more filings, less per-filing signal, and a much louder public stage. If you want a quiet edge, you have to build it before the stage gets loud.

The bigger pattern

Cal-WARN was designed in a world where layoffs were episodic and industrial. It is being asked to monitor a labor-market shift that is continuous, white-collar, and disproportionately concentrated in a state that hosts the companies driving the shift. The 180 days is not really about WARN. It is about whether California can build a regulatory feedback loop fast enough to keep up with a workforce restructure that, by the numbers above, is already running 33 percent ahead of last year.

For sourcers, the meta-lesson is the one every previous filing-driven advantage has taught: signals get commoditized the moment they get a dashboard. The work that survives is the work upstream of the signal: knowing the people, the teams, the GitHub histories, the company-internal politics that explain why a particular engineer is reachable this quarter and not next. That work does not get easier in November. It gets more valuable.

The Burlingame 124 are still findable today. So are the Sunnyvale 74. So is the next filing that hits while you are reading this. The clock started on May 21.

FAQ

Does the executive order change what I can do as a sourcer today?

No. EO N-6-26 directs LWDA and EDD to study and recommend. It creates no new obligation on private employers and changes none of the existing Cal-WARN thresholds (75-employee, 50-layoff, 60-day notice). Today's filings flow under the same rules they did on May 20. The change comes after the mid-November recommendations are adopted in some form, likely through SB 951 or a successor statute.

When does the EDD AI-jobs dashboard go live, and should I worry about it?

Within 90 days of May 21, 2026, so roughly mid-August. You should not worry about it, but you should plan around it. The moment it ships, the company-and-sector-level signal you currently extract from WARN filings becomes public and universal. Your edge has to shift from "I read the filing first" to roster quality, outreach timing, and below-WARN coverage. That is where tools like Refolk earn their keep, because the comparative advantage moves from data access to candidate ranking.

Why is WARN missing so many AI engineers?

The thresholds. Cal-WARN only triggers at 75-plus employee employers with 50-plus affected workers, with 60 days notice. A Series B that cuts a six-person AI team never files, and individual terminations never file. Stanford HAI's 2026 AI Index shows employment for 22 to 25-year-old developers down nearly 20 percent from 2024, much of it concentrated at small employers. If your pipeline only watches WARN, you are systematically blind to the most interesting slice of the displaced pool.

What should I do in the next 30 days specifically?

Three things. One, set up named-employer alerts on WarnTracker or WARN Firehose for every target company in your patch, and pipe them into Slack or your ATS. Two, build location-level rosters for the top 20 employers you'd recruit from, so you are matching against pre-built lists when filings drop instead of searching cold. Three, set up a parallel below-WARN watch: GitHub activity, company About-page diffs, and plain-English candidate queries through Refolk for the startup-scale cuts that never make a filing.