The AI Disclosure Gap, May 18: 113,863 Tech Layoffs and No Federal Law Says Companies Have to Tell You Whether AI Did It — PREPARE Act (S.3339) Stalled in Senate, Colorado AI Act Paused by Federal Court April 27 and Replaced by SB 189, California SB 951 90-Day Notice Bill Still in Committee — Cisco's CFO Insists the Cuts Are 「Not Savings-Driven」 and No Worker Has Any Way to Audit That | LostJobs.AI

As of May 18 the U.S. tech-sector layoff tally for 2026 is 113,863 workers across 179 distinct events. Challenger, Gray & Christmas put the first-four-month total at 85,411 — a 33% jump on Q1 2025 and the worst start to a year since 2023. The headline number that lands with all the cuts is 825 layoffs per calendar day, every day, for nearly five months.

Not one of those 113,863 workers had any federal right to know whether AI actually caused the cut.

That is not a rhetorical line. The Worker Adjustment and Retraining Notification Act — the only federal statute that touches mass layoffs — contains no provision requiring an employer to say whether an AI system replaced the role, which system, what its accuracy was, or whether retraining was offered first. A company can announce 30,000 cuts, cite “AI productivity” in the press release, and never have to back any of it up. As of May 18, 2026, every part of that sentence is still true.

The bills that would change that, and where they are

Two federal bills sit in committee.

The AI Workforce PREPARE Act (S.3339) — introduced in the Senate in October 2025 — would amend the WARN Act to require companies to specify when AI was a substantial factor in a mass layoff, name the AI systems used, estimate the percentage of job losses attributable to them, and describe the retraining offered before cuts were made. The bill has had zero committee actions since referral. The House companion, the No Robot Bosses Act, would require human oversight and disclosure whenever AI tools are used in employment decisions. Same status.

Eight months in committee. Across that window, the cohort tracked by Challenger has added something north of 100,000 cuts. The cost of the disclosure regime not existing is, on a 2026 run rate, roughly 200,000 workers per year whose stated reason for being out of work cannot be audited.

State preemption has not bailed the federal level out either. Colorado’s Artificial Intelligence Act was supposed to take effect June 30, 2026 — six weeks from today — with a duty of reasonable care against algorithmic discrimination in employment decisions. On April 27, 2026 a federal court paused enforcement, and the state legislature is set to replace the law with SB 189, described by Senate Majority Leader Robert Rodriguez as “more of a notice bill.” The original disclosure teeth do not return in the new draft.

California SB 951 — the Worker Technological Displacement Act — would require 90 days’ notice before AI-driven layoffs with disclosure of the AI systems used. Still in committee. Illinois and New York City require disclosure of AI use in hiring tools, but enforcement has been close to zero.

The map is consistent: every bill written to close the disclosure gap is currently either in committee, paused, or being watered down.

The cohort the regulators are watching, layoff by layoff

Date	Company	Cuts	Stated cause	Auditable?
Jan 2026	Nike	~800 distribution	”accelerate automation”	No
Q1 2026	Salesforce	undisclosed AI tranche	”Agentforce productivity”	No
Mar 2026	Oracle	up to 30,000 (~20%)	AI data-centre pivot	No
May 5	Coinbase	700 (~14%)	“AI-augmented”	No
May 5	PayPal	up to 4,760 (~20%)	AI + $1.5B savings	No
May 7	Cloudflare	1,100 (~20%)	“agentic AI era”	Partial — internal usage cited
May 11	GM IT	600	swap conventional eng for AI-native	No
May 13	Cisco	4,000 (<5%)	“not savings-driven, reallocate to AI”	No — orders book shown, attribution not
May 13	LinkedIn	875 (~5%)	Roslansky: explicitly “not AI”	No
May 14	Kraken / Payward	150 (~5%)	internal AI deployment	No
May 15	Innovaccer	340	”AI-native transition”	No
May 15	Starbucks	300 (third Niccol round)	operating-model efficiency	No
May 20 (planned)	Meta	8,000 (~10%)	$125–145B AI capex	No

The only column that matters is the right one. Every row asserts an AI cause. Not one comes with the underlying data that would let a journalist, an academic, or a former employee verify the claim. The cohort’s stock-price punishment is consistent with that opacity: the market is repeatedly pricing the press release at less than the cohort’s CFOs do, and the most parsimonious explanation is that nobody outside the company can tell whether the AI savings are real until two earnings cycles later.

The most-quoted skeptics, in their own framing

Wharton management professor Peter Cappelli — who has spent four decades studying corporate workforce decisions — put the gap as plainly as anybody this cycle: “The headline is, ‘It’s because of AI,’ but if you read what they actually say, they say, ‘We expect that AI will cover this work.’ Hadn’t done it. They’re just hoping.”

Sam Altman, who has reason to want AI displacement to be real and provable, conceded in February that “some AI washing where people are blaming AI for layoffs they would otherwise do” is part of the cycle, while also confirming that genuine displacement is occurring. The two are, from the outside, indistinguishable.

Oxford Economics concluded in January that firms “don’t appear to be replacing workers with AI on a significant scale.” Deutsche Bank wrote in its January year-ahead that “AI redundancy washing will be a significant feature of 2026.” Andy Challenger, the lead voice on the only public layoff dataset most reporters cite, has been more careful: “regardless of whether individual jobs are being replaced by AI, the money for those roles is.”

The unifying point is not that AI never causes a layoff. It is that on the public evidence, in May 2026, no outsider can reliably tell which 30% of the cohort is which.

Cisco’s “not savings-driven” claim is the test case nobody is allowed to grade

The cleanest case study is also the loudest. On May 14, Cisco notified roughly 4,000 workers of termination on the same day it reported record Q3 FY2026 revenue of $15.8B. CFO Mark Patterson told analysts the restructuring was “not a savings-driven exercise” but rather a reallocation of capital toward silicon, optics, security, and AI — a claim backed by $5.3B in YTD AI infrastructure orders and a raised FY26 AI orders forecast of $9B, up from a prior $5B.

The orders book is auditable. The attribution between the orders book and the 4,000 cuts is not. Cisco was not required to disclose — and did not disclose — which job functions the AI orders displaced, what the headcount path looks like in the silicon, optics and security organisations that are supposedly absorbing the reallocation, or what the retention rate is on workers who were offered internal moves rather than termination. The PREPARE Act would have required all three. The PREPARE Act is in committee.

The human side of the disclosure gap, in two numbers

Tech-sector unemployment hit 5.8% in early 2026 — the highest level since the dot-com bust of 2001–2002. The headline U.S. unemployment rate sits at 3.8%. Tech is now running two points hot to the broader labour market for the first time since 2003.

The median time to next role for a laid-off tech worker has stretched from 3.2 months in 2024 to 4.7 months in 2026. A 2026 Motion Recruitment study finds AI adoption is slowing hiring for entry-level and generalised IT roles while demand for AI-engineering roles surges. A Stanford study published in late 2025 documented a 16% relative employment decline for recent graduates in AI-exposed roles versus stable employment for experienced workers — the class-of-2026 squeeze we covered May 17 is the same data series.

Daniel Zhao, chief economist at Glassdoor, captured the cycle’s defining shift: “Because natural attrition isn’t happening as much, companies are being more aggressive about pushing people out the door.”

The gap that matters here is asymmetric. The companies have full visibility into whether AI is doing the work. The displaced worker, the hiring manager at the next firm, the journalist writing the story, and the academic running the dataset have none.

What to watch

The PREPARE Act’s first committee action. Eight months of nothing is the actual story. Any movement before end-Q2 would be the first time federal disclosure caught up to a cohort already at six figures.
Colorado SB 189’s final text. “Notice bill” is doing a lot of work in that description. Whether the law preserves any of the original AI Act’s algorithmic-discrimination teeth, or strips them entirely, sets the template the other states will copy.
The next CFO who says “not savings-driven.” Cisco was the first to thread that needle on the same day as record revenue. If two more do it in Q2 earnings, the disclosure regime — voluntary, narrative, unverifiable — is the regime.
A Meta retention-rate disclosure. Meta cuts 8,000 on Wednesday May 20. If the company quietly tells the Street six months from now what percentage of the 8,000 took internal AI-pod roles versus exited the company, the disclosure gap closes one company at a time. If it does not, the cycle’s biggest single number stays a black box.

The version of this story the corporate communications teams want is that AI is doing the work and 113,863 workers are the necessary friction of a real productivity shock. The version the bills in committee would force is the one where the companies have to show their work. As of May 18, the second version is still illegal to demand.