California's No Robo Bosses Act: what a real human review would look like

If you run HR alone, some of your discipline decisions already start with software. An attendance system flags late punches. A productivity dashboard marks someone as below target. A scheduling tool counts missed shifts. You read the flag, you talk to the manager, and a warning goes out.

California’s legislature has now passed a bill about exactly that moment. It is not law yet. Four bills on AI at work are waiting for Governor Newsom, who has until September 30 to sign or veto them. But the most relevant one for a solo HR lead, SB 947, the “No Robo Bosses Act,” is worth understanding now, because it describes a habit any careful HR team can adopt whatever the governor decides.

This post is general education, not legal advice.

What SB 947 would require

According to Mintz’s summary in the National Law Review, employers “may not rely solely on an ADS when making a disciplinary or termination decision.” ADS means automated decision system. If an employer primarily relies on one for discipline or firing, a person would have to corroborate the decision using the data the system collected or other relevant information.

The affected employee would also get a written notice when the decision is communicated. Hoodline reports that the notice would include a statement that the worker will not face retaliation for using their rights, contact details for the human reviewer, and the right to see a description of the data the system used.

Two details matter for small employers:

• The definition is broad. Hoodline quotes the bill’s definition of an automated decision system as “any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence used to assist or replace human discretionary decisionmaking.” Depending on how it is read, that may reach ordinary HR software, not only tools marketed as AI.
• The dates. If signed, it takes effect July 1, 2027. Hoodline reports penalties of $500 per violation.

The other three bills are narrower for most small employers. SB 951 would require layoff notices under California’s WARN rules to say when a mass layoff is caused by AI. AB 1883 would bar AI surveillance tools that infer employees’ emotions or collect neural data. AB 1331 would bar workplace surveillance tools in bathrooms, with narrow exceptions.

A worked example

Take a 50-person distribution company in California with one HR manager. The timekeeping system produces a weekly exceptions report and suggests a written warning for anyone with five or more late punches in a month. This month it flags a warehouse lead.

Here is what relying on the tool looks like: the report says five, the template fills itself, the warning goes out.

Here is what corroboration could look like, done by the HR manager in about twenty minutes:

1. Pull the raw records, not the summary. Open the actual punches for the month. Two of the five were eight minutes late on a day the badge reader was replaced. That is a data problem, not a conduct problem.
2. Check what the system cannot see. The employee had an approved schedule change for one week that was never entered in the timekeeping tool. One more flag falls away.
3. Ask the person. A short conversation surfaces a childcare change that explains the remaining two. That may raise a leave or accommodation question, which is a separate conversation with its own rules.
4. Decide, and write down why. Two late punches remain. The manager decides on a verbal reminder, not a written warning, and records what they checked and what changed their view.

Notice what happened. The tool was useful: it found a pattern worth a look. The decision was different from the one it suggested, and the HR manager can explain every step. That is the difference between reviewing a decision and signing it.

A corroboration record you can reuse

A one-page note for any discipline or termination decision where software produced the first signal:

• Signal: which tool, what it flagged, on what date.
• Source data checked: the raw records opened, not the dashboard.
• Gaps: what the system could not know (approved changes, leave, equipment faults).
• Employee’s account: when you spoke and what they said.
• Decision and reason: what you decided, in your words, and whether it matches the tool’s suggestion.
• Reviewer: your name, the date, and who else was consulted.
• Notice: what the employee was told and when.

Keep it with the personnel file. Whether or not SB 947 becomes law, this is the record you would want if the decision is ever questioned.

What to do this week

• List every tool that produces a flag, score or ranking that has ever fed into a warning, a performance plan or a termination. Include timekeeping, scheduling and sales dashboards.
• Name the person who reviews each one before anything reaches an employee. If the answer is “nobody, it goes straight out,” start there.
• Try the corroboration record on the next flagged case, even a small one, and time it.
• Watch for the governor’s decision by September 30, and if you employ people in California, ask your employment counsel what it would mean for your notices and records.

The software will keep getting better at finding patterns. The part that stays with you is deciding what the pattern means for a person, and being able to say why.