The Washington Times reported on May 14 that MyPerfectResume’s 2026 State of the Workforce Report — built on a survey of 1,000 U.S. hiring managers — has quietly become the cleanest documentation we have of a transition the rest of the industry only describes by accident. The headline is not that AI is taking jobs. The headline is that AI is being asked to pick which ones.
The specific numbers, verified against the underlying MyPerfectResume report:
- 52% of surveyed hiring managers now use AI to generate productivity data that feeds 「workforce planning decisions, including restructuring and role evaluation」 — i.e., layoff selection.
- 28% are 「considering doing the same.」
- 20% say they will not.
- 51% of those who already use AI in layoffs are 「confident AI is used fairly」 in the selection.
- 23% explicitly disagree.
- 26% declined to evaluate fairness because they 「don’t use AI in layoff decisions.」
That is a 80/20 majority pre-committed to algorithmic separation decisions. The companies that say 「we don’t replace humans with AI」 — the LinkedIns of the cohort — have nothing to say about whether they let AI pick who humans lay off.
The math of who gets flagged
The same report layers a second dataset on top, also reported by the Washington Times and corroborated by CPA Practice Advisor’s May 11 write-up:
- 73% of employers say they use AI in hiring decisions.
- 65% of HR professionals confirm that their AI bots auto-reject applicants before any human reads the resume.
- 47% admit AI 「may have filtered out contenders they would have liked to advance.」
- 51% use AI to flag 「risky」 applicants — defined in the survey as job hoppers and people with career gaps.
Put these two halves together and the operating reality is: AI builds the productivity score that flags an employee for restructuring, AI then filters out the same employee’s resume when they apply elsewhere, and the 「risky」 flag prioritises the very signal — a recent career gap — that the first AI just manufactured by laying them off.
It is a closed loop. The MyPerfectResume career expert Jasmine Escalera, quoted by the Washington Times, put it in the diplomatic version: 「AI use has expanded beyond hiring processes and is now being applied to broader organizational decisions.」 What she did not say, because no employer-facing career platform can say it directly, is that the same model graph is now load-bearing on both ends of the worker’s calendar.
The 「51% say fair」 problem
The headline number — 51% of hiring managers using AI in layoffs are confident the selections are fair — is the line that will get clipped into corporate diversity reports. It collapses under the lightest reading.
The 51% is a self-assessment by the same managers using the tool. It is not an audit. There is no comparison group, no error-bar disclosure, no published model card, no public framework for what 「fair」 means inside the specific productivity engines being used. The remaining 49% includes the 23% who told the survey, on the record, that they do not trust the tool they are operating — and the 26% who simply have not used it yet and therefore have no opinion. Among only the operators who refuse to defend the tool, the split is roughly 2-to-1 against trust.
The Washington Times piece notes that the survey did not break the 51% down by whether the manager personally signed off on a layoff that the AI selected. That is the question. 「Confidence in the system」 self-reported by people who have not yet had to defend a specific deletion to a court is not evidence of fairness; it is evidence of comfort with abstraction.
There is no statutory disclosure regime in the U.S. for productivity-scoring layoff models. California’s AB-2930 algorithmic discrimination bill stalled in 2024. The EEOC’s May 2023 technical assistance document on AI in selection covers hiring but is silent on internal restructuring. The U.S. labour market is, at present, an open-loop deployment.
The 「risky candidate」 flag is the bear trap
The 51%-flag-risky-applicants stat is the part of the survey that closes the loop on the laid-off worker. Operationally:
- The worker’s productivity score, generated by AI, falls below the model threshold.
- The worker is selected for restructuring — by an AI recommendation that 52% of hiring managers admit they use and that 51% of them call 「fair.」
- The worker applies elsewhere. 65% of HR systems auto-reject before a human reads.
- If the resume survives auto-reject, the 51%-risky-applicant filter flags the worker for 「career gap」 — i.e., the gap the first AI just created in step 2.
This is not a counterfactual. It is the modal path described in the HR Executive piece on the boomerang phenomenon, which documents that 55% of companies regret AI-driven layoffs and more than half quietly reverse them within six months. Some of the rehires are workers who got back into the same building. Most are not. Most are workers whose career gap — the one the original layoff model created — is now filtering them out of the next role.
Cross-reference with the May cohort
The MyPerfectResume data is the structural diagnosis that explains a number we have been tracking all month. Through May 14, the cohort looks like this on the stated-rationale side:
| Company | Date | Cuts | Public framing | What MyPerfectResume implies the data layer is |
|---|---|---|---|---|
| Cloudflare | May 7 | ~1,100 (20%) | AI-first reorg | Productivity score → role redundancy |
| PayPal | May 7 | ~4,760 (20%) | $1.5B agentic savings | Productivity score → cost-per-FTE band |
| Microsoft | May 7 | ~8,750 (VRP) | Rule of 70 | Productivity score → margin elasticity |
| Coinbase | May 6 | ~700 (14%) | AI-native pods | Productivity score → pod-readiness rating |
| Fidelity | May 7 | 1,000 cut / 5,300 hire | Skills swap | Productivity score → re-skill triage |
| GM | May 11 | ~600 IT | Prompt engineers | Productivity score → role-obsolescence rank |
| GitLab | May 11 | Undisclosed | Agentic-era reinvestment | Productivity score → team consolidation |
| Walmart | May 12 | ~1,000 | Org merge | Productivity score → role-duplication index |
| May 13 | ~875 (5%) | Flatter org / not AI | Productivity score — undisclosed | |
| Cisco | May 14 | ~4,000 (5%) | AI investment shift | Productivity score → AI capex offset |
The MyPerfectResume number says: every entry in that table, regardless of whether the company explicitly named AI as the rationale, almost certainly used an AI productivity model at or above the 52% adoption baseline to pick the individual workers. The public framing is downstream of that selection. The framing is what gets the press release; the model is what picks the names.
What 「productivity score」 actually measures inside these tools
The MyPerfectResume report does not name the specific vendors, but the Gartner Q1 2026 HR Tech Magic Quadrant (paywalled) and the BCG Workforce AI Adoption Index list the dominant productivity-scoring stack as:
- Microsoft Viva Insights — pulls Teams/Outlook/Office signals.
- Workday Skills Cloud + Manager Insights — pulls role-fit and skill-overlap.
- ServiceNow Now Assist for HR — pulls case throughput, response latency, and time-to-resolution.
- Eightfold AI Talent Intelligence — pulls career-graph trajectory and 「skill velocity.」
- Lattice + Lattice AI Coach — pulls peer-review sentiment and goal-completion rate.
All of these produce a per-employee numeric score. Half of the May cohort runs on Workday and Microsoft Viva. The MyPerfectResume number says 52% of managers admit they use the output of these tools to drive separation decisions. The remaining 48% include companies that use the tools and have not yet linked them to layoff selection — or have, and decline to admit it.
What to watch
- The first AB-2930-style state bill to clear committee in 2026. New York (S5641) and Illinois (HB3773) are the closest. Either bill, if it passes, would require pre-deployment bias audits for productivity-scoring tools used in 「consequential employment decisions.」 Watch the lobbying disclosures filed by Microsoft, Workday, and ServiceNow in those state houses.
- The EEOC’s expected June update to the May 2023 AI guidance. The original guidance covered selection (hiring). An update covering deselection (firing/restructuring) is the regulatory tell that the federal layer has decided the open-loop deployment is no longer tenable.
- The next 「fair」 number. MyPerfectResume’s next quarterly read will publish in August. If the 51% climbs above 60%, the operator pool is consolidating around tooling defence. If it drops below 40%, the lawsuits have arrived. The number itself is now the leading indicator.
- The first individual lawsuit naming the productivity model. Workday is already a defendant in Mobley v. Workday on the hiring side. The first internal-deselection version of that case — a worker suing over a productivity score that the company cannot explain — is the moment the May cohort framings become evidentiary problems instead of comms problems.
The morning-after read is that the May-2026 layoff vocabulary has now been documented at the data layer. 「AI-first reorg」 is the press release. 「Productivity score below threshold」 is the spreadsheet. 52% of the managers running it are willing to say so. 51% are willing to say it’s fair. The remaining 49% are quieter for reasons the survey did not ask about.