Governor Gavin Newsom has until September 30 to sign or veto California's SB 947, the "No Robo Bosses Act." Most of the coverage has been about big tech employers and warehouse robots. That's the wrong frame.
The real story is smaller and closer to home: the bill covers any employer who uses an automated decision system (ADS) when disciplining or terminating workers. The definition of ADS in the bill is broad — "computational processes derived from machine learning, statistical modeling, data analytics, or artificial intelligence techniques." That's not future science fiction. That's the AI performance dashboard your HR lead checked before the last PIP conversation.
If Newsom signs it, SB 947 becomes operative on July 1, 2027. The penalties are $500 per violation, plus the ability for workers to pursue punitive damages and attorneys' fees in private lawsuits. The clock starts counting a few quarters from now — which is enough time to prepare, but not enough time to ignore.
What the Law Actually Says
SB 947 doesn't ban AI from HR decisions. It bans relying solely on AI for termination or disciplinary decisions.
Here's the key sentence: if an employer "primarily relies" on ADS output for a discipline or termination decision, a human must independently corroborate that output using relevant information. And if the output can't be corroborated — or is found to be inaccurate, incomplete, or misleading — it cannot be used.
That sounds reasonable. The problem is that a lot of small orgs have implemented HR tools specifically because they reduce the time humans spend on these decisions. They've built workflows where someone reviews a dashboard, sees a low score, and proceeds. That might not count as genuine independent corroboration.
The question that will determine compliance: is a human actually making an independent judgment, or are they signing off on what the system recommended?
The Part Your HR Lead Hasn't Noticed
Modern HR software comes with AI features baked in. Not as an optional add-on. As the default experience.
Lattice generates performance scores and surfaces employees it identifies as flight risks or underperformers. Rippling runs algorithmic alerts on performance and attendance metrics. Culture Amp uses AI to flag engagement patterns. BambooHR's performance management module generates ratings-based recommendations. These aren't fringe tools — they're what a 15-50 person org uses for people management.
None of these tools were sold as "the AI that fires your employees." They were sold as tools to save HR time and make more consistent decisions. That's the quiet part of this law: the scope isn't limited to obvious AI systems. It includes any workflow where an algorithm's output is a primary input to a discipline or termination decision.
Worth noting: the bill covers independent contractors, not just employees. If your org works with a large freelance base and you use automated performance scoring to decide who gets more work (or less), that's potentially in scope.
This Isn't Just California
If Newsom vetoes SB 947 again — he vetoed the original version in October 2025, citing overly broad language — that doesn't make this problem go away. The 2026 version was revised specifically to address those concerns, passing 53-14 in the Assembly and 28-10 in the Senate. The legislative momentum is clear.
And the regulatory landscape outside California is already moving:
- NYC Local Law 144 already requires employers using AI in hiring to conduct bias audits and disclose AI use to candidates. It's been in effect since 2023.
- Colorado's Automated Decision-Making Tools (ADMT) Act covers high-risk AI decisions in employment, among other sectors.
- The EU AI Act's employment provisions prohibit high-risk AI systems in hiring and performance management without specific transparency and human oversight requirements.
The pattern is consistent: jurisdictions are moving toward requiring a documented human decision layer over AI-assisted HR tools. California SB 947 would be the first US state mandate specifically for discipline and termination, but it's the most explicit articulation of a direction everyone is heading.
If you have workers or contractors in any of these jurisdictions — or plan to — this framework is coming for your HR stack eventually.
What to Do This Week
You don't need to rebuild anything before September 30. But you should know what you're dealing with.
Audit your HR tools for AI features. Log into Lattice, BambooHR, Rippling, Culture Amp, or whatever you use and look for performance scoring, risk flagging, or recommendation features. Document which decisions those features inform.
Check your documentation trail. If you put someone on a PIP or terminated an employee in the past year, is there a written record of the manager's independent assessment — separate from the software's output? That paper trail is what compliance looks like under SB 947.
Talk to your HR leads about what "human review" actually means. There's a difference between a manager who reviews AI output and genuinely exercises independent judgment, versus one who sees a low score and clicks confirm. Build a process where the manager's reasoning is documented independently of the tool's recommendation.
Check vendor settings and disclosures. Some HR platforms have documentation on how their AI features work and whether they provide outputs that are "recommendations" versus "decisions." That distinction will matter in a compliance conversation.
For orgs outside California: The compliance question isn't whether this bill passes on September 30. It's whether you have a documented human decision layer in your HR process. If the answer is no, you're building technical debt that will catch up with you regardless of jurisdiction.
The Bottom Line
The "robo boss" framing in the bill's name makes this feel like a problem for Amazon warehouse operators or algorithmic gig platforms. It isn't. It's a problem for any small org running modern HR software and assuming that "AI-assisted" means someone reviewed it.
The compliance requirement is genuinely achievable: make sure a human is making an independent call, and document it. The organizations that will have problems are the ones where HR is one person, the software surfaces a recommendation, and there's no record that anyone actually made a decision.
This is the kind of process question we work through with teams in a half-day sprint — what decisions is your AI stack actually making, what does your documentation trail look like, and where are the gaps. If that's a conversation worth having before July 2027, reach out.