California Just Regulated AI Firing Decisions. It Left Hiring Alone

On September 30, 2026, Governor Gavin Newsom signed SB 947, the "No Robo Bosses Act," reversing his own veto of an earlier version of the bill just eleven months earlier. Coverage called it a landmark — the first state law to tell employers they cannot let an algorithm fire or discipline a worker on its own. That description is accurate. It is also, for anyone reading the headline as "California now regulates AI hiring and firing," wrong in a way that matters.

SB 947 regulates the back end of employment. It says nothing about the front end — the resume screens, the AI interviews, the candidate-scoring tools that are where most AI adoption in HR is actually concentrated right now. That split wasn't an oversight. It was a deliberate rewrite, and it tells you something about where AI-in-employment law is heading next and where it isn't — yet.

What SB 947 actually does

The bill, authored by state senator Jerry McNerney, prohibits California employers from relying solely on an "automated decision system" (ADS) — defined broadly as a computational process built on machine learning, statistical modeling, or AI that replaces human discretion in producing a score, recommendation, or decision affecting a worker — to discipline or terminate an employee. If an employer's decision primarily rests on an ADS output, a human reviewer has to corroborate it against independent evidence: manager evaluations, personnel files, peer reviews, witness interviews. If that corroboration can't be done, or the human reviewer concludes the ADS output is wrong, incomplete, or misleading, the employer cannot act on it.

Workers disciplined or fired under a decision that primarily relied on an ADS must get written notice: that an automated system was involved, what data it drew on, and a named human they can talk to about the decision. Enforcement runs through the state labor commissioner and public prosecutors, with civil penalties of $500 per violation stacking on top of potential punitive damages and attorney's fees. The operative date is July 1, 2027, giving employers roughly nine months to get their termination and discipline workflows in order.

Newsom signed it alongside two related bills the same day: AB 1883, which bans workplace-surveillance tools that infer a worker's emotional state or collect neural data and requires employers to report AI-involved layoff decisions to the state, and AB 1331, which bars AI-driven monitoring in bathrooms. Reporting on the day — CNBC, KQED, and the governor's own office — framed all three as a single "worker protections" package, which is fair: together they're the most specific statutory guardrails any state has put on AI's role in managing an existing workforce.

The provision that didn't survive

The bill that got signed is narrower than the bill that got vetoed. The 2025 version — then numbered SB 7 — would have required employers to notify every worker "foreseeably" affected by an ADS, maintain a running inventory of every ADS in use, cover gig workers and "deactivation" decisions (the language that drew Uber and Lyft's lobbying attention), and, notably, tell job applicants when an automated system was used in a hiring decision. Newsom's veto message called that version "unfocused" and its notice requirements "overly broad."

When McNerney reintroduced the bill, the rewrite didn't just narrow the notice requirements he'd been criticized for — it dropped hiring from the bill's scope entirely, along with the gig-worker and deactivation language. SB 947 as signed speaks only of "employees" and of "disciplinary or termination" decisions. A job applicant screened, scored, or rejected by an AI tool before they're ever hired gets none of the corroboration right, the notice right, or the named-human-contact right that a current employee facing AI-assisted discipline now gets under California law.

That's a real asymmetry, and it's worth sitting with rather than treating as a footnote. Most of the AI investment HR teams are making right now sits on the hiring side of that line — resume vetting, screening interviews, candidate scoring against structured pipeline stages — not on the discipline-and-termination side the new law actually reaches.

Why hiring is the harder thing to legislate — and why that's not an accident

There's a structural reason the bill's scope shrank to termination, not just a political one. A termination decision has a clean shape for a statute to grab onto: one moment, one worker, one outcome, usually with an existing body of wrongful-termination law to attach new procedural requirements to. A hiring decision doesn't have that shape. It's a funnel — resume parsing, an initial AI screen, maybe a scored interview, maybe a human-led final round — applied simultaneously to hundreds of candidates, most of whom never get a single identifiable "decision event" to attach a notice-and-corroboration requirement to. Writing "a human must corroborate the termination" is one sentence. Writing the equivalent for "a human must corroborate every point in a hiring funnel where an ADS influenced who advanced" is a much harder drafting problem, and it shows in how fast the hiring-notice language was the first thing cut when the bill needed to get narrower to get signed.

The political economy points the same way. "Don't let a bot fire you with no one checking" is a protection for workers who already have the job — a constituency with an obvious, sympathetic story and no organized opposition eager to defend robo-firing. "Slow down the AI tools screening who gets the job in the first place" runs directly into the HR-tech vendor industry, employer efficiency arguments, and the much blurrier question of what "corroboration" would even mean applied to a screening funnel instead of a single termination call. It's not a coincidence that the bill which survived two legislative sessions is the one with the cleaner shape and the narrower set of people pushing back.

The federal backdrop makes the gap bigger, not smaller

This isn't happening against a stable federal baseline. The EEOC pulled its AI-in-employment technical assistance documents from eeoc.gov in January 2025 following a change in administration priorities, and as the National Law Review has reported, that guidance has stayed down for well over a year, with the current EEOC leadership not listing AI-in-hiring among its enforcement priorities. States have been filling that vacuum unevenly — Illinois and Colorado both have AI-employment statutes in some stage of effect or delay, New York City's Local Law 144 has required independent bias audits of "automated employment decision tools" used on NYC-based candidates since 2023, and Connecticut's new AI law (effective October 1, 2026) covers a different slice again, built around whistleblower protections at AI developers rather than employer use.

Put plainly: there is no single federal floor right now, the state floor is a patchwork of differently-defined terms, and California — the state most willing to legislate aggressively on AI surveillance and AI-driven termination in the same week — chose not to extend its flagship 2026 bill to hiring at all. If you are a hiring team concluding "it's fine, no one has written the hiring-specific rule yet," you have correctly read the current statute and incorrectly read the trend. A provision getting cut from a bill under lobbying pressure in one legislative session is not the same as a legislature deciding, on the merits, that hiring doesn't need the same scrutiny it just gave firing. McNerney's office has not signaled the fight over the applicant-notice language is closed for good, and the asymmetry between "we just regulated the exit" and "we did nothing about the entrance" is exactly the kind of gap that tends to get closed in the next session once a few AI-hiring incidents get litigated under ordinary disparate-impact law in the meantime — law that never went away just because there's no AI-specific hiring statute yet.

What this means if you're actually running a hiring pipeline today

A few things worth doing now, independent of what any statute currently requires:

Don't wait for a hiring-specific corroboration law to build corroboration into your hiring process. If SB 947's logic — a human has to be able to point to what else supports an AI-influenced call — ever gets extended to hiring, the employers already keeping that record will have a much shorter runway to comply than the ones starting from zero.
Keep a plain record of what criteria were applied to a candidate, and apply them consistently. This is table stakes for defending any hiring decision under existing discrimination law, AI-assisted or not, and it's the same discipline SB 947 just made mandatory on the termination side. NiceHire's own screening pipeline, for example, scores every candidate against the same fixed criteria — technical fit, communication, cultural fit, each out of 100 against a configurable threshold — specifically so there's a consistent, documented basis for how a candidate was evaluated. That's a claim about the mechanism, not about whether any given outcome was fair; no vendor, including us, can credibly self-certify that.
If any interview or screening tool you use infers emotion or reads biometric/neural signals, treat AB 1883's logic as a preview, not a California-only quirk. The bathroom-monitoring bill that passed alongside it shows how far this legislature is willing to go on AI surveillance; emotion-inference bans on the hiring side are a plausible next target even though AB 1883 itself is written around workplace surveillance of existing employees.
Track this at the legislative level, not just the compliance level. The hiring-notice provision was cut for drafting and political reasons this round, not abandoned on principle. Treat "no hiring-specific AI law yet" as a current fact, not a settled one.

Sources: CNBC, KQED, HR Dive, Ogletree Deakins, Crowell & Moring, National Law Review.