AI layoffs need an automation map

In short: California’s AI-layoff notice rule turns automation into an attribution problem; map systems to job functions before decisions harden.

California approved SB 951 on 30 September 2026. Its chaptered text keeps the existing 60-day Cal/WARN notice and adds specific fields when a covered mass layoff, relocation or termination is caused in whole or substantial part by AI or automated technology.

That makes AI layoffs an attribution problem before they become a notice problem. My claim is simple: an organisation needs an automation map that joins deployed systems to job functions and workforce decisions while the rollout is happening. Reconstructing that chain after positions disappear will invite a confident story where evidence should be.

SB 951 asks what the automation changed

The chaptered SB 951 text applies the additional disclosure inside California's existing WARN framework. It defines a covered establishment as one employing, or having employed in the preceding 12 months, at least 75 people. A mass layoff means 50 or more employees in 30 days.

For a covered action linked to AI, the notice must identify the affected number and occupation or classification, work location, the job functions being automated and the category or type of system. It also puts “This notice is for a technology displacement” at the top. The governor's signing announcement describes the measure as workplace transparency, not a ban on automation.

This article is not a coverage opinion for a particular employer. The causation phrase — “in whole or in substantial part” — is exactly where a qualified legal assessment will matter. Engineering still has a job before that assessment starts: preserve the factual deployment trail.

The chaptered law does not name the vendor

The mailbox analysis that led me here said the notice must name the entity that developed, sold or leased the AI product. An earlier amended draft did contain that field. The chaptered law does not. It requires the category or type of system.

That correction is not a footnote. It is a small demonstration of the same operational risk: a plausible summary can retain a requirement after the final text removed it. I made a related argument about versioning statute calendars. Here the sharper rule is to pin the chaptered source beside the control derived from it. A policy built from the wrong draft can be beautifully implemented and still answer the wrong law.

The automation map I would require

I have not handled a California AI-linked layoff for Trigent or a client. This is my editorial judgment about the operating record I would want engineering, operations, HR and counsel to review together.

For each material automation, record the job function touched, the human work removed or changed, the deployed system and version, rollout date, accountable owner, evidence used in the workforce decision and later changes to that decision. Do not label every productivity tool as displacement. Do not wait for HR to infer causation from a procurement invoice.

The useful unit is a job function, not a product licence. One assistant may draft reports, classify tickets and prepare orders across three teams. The map should show which capability changed which work, and when. That gives counsel evidence to assess; it does not automate the legal conclusion.

EDD reporting will turn notices into a public dataset

SB 951 requires California's Employment Development Department to publish notice summaries and a quarterly statewide summary of reported technology displacement. The current EDD WARN guide explains the 75-person, 50-person and 60-day baseline but had not yet incorporated the new AI fields when I checked it on 7 October.

That means implementation guidance can still mature. The chaptered text is the present anchor. The automation map is the internal evidence layer that should survive later form and guidance changes.

What's in it for you

Engineering can explain what capability changed instead of handing HR a vendor list.
HR and counsel can test a causation narrative against dates, functions and decision evidence.
Leaders can correct source-version drift before it becomes a control or a public notice.

Map the automated job function when you deploy the system, not when you are forced to explain the layoff.