California's New AI Laws: Human Handoff and No Robo Bosses

If you run a California business, two new AI laws matter to you. SB 947, the No Robo Bosses Act, applies to essentially every California employer: starting July 1, 2027, you cannot let an automated system be the only thing behind a decision to discipline or fire someone. AB 1609, the Right to Human Customer Service Act, legally covers only companies with more than $500 million in annual revenue, but it sets the standard customers will now expect from every chatbot and AI receptionist: say it's a bot, and give people a quick way to reach a human.

So the short answer for a 12-person clinic in Irvine or a 40-person trades company in the Inland Empire is: SB 947 binds you, AB 1609 probably doesn't (yet), and the cheapest way to handle both is to design a human into the loop now, before you buy or build anything else.

Below is what Governor Newsom actually signed, in plain English, what each law asks of you, and a checklist you can work through before mid-2027.

What Newsom signed, in plain English

In the last week of September the governor signed a package of AI bills. The ones a Southern California owner should know:

The governor's Sept. 28 announcement describes AB 1609 as requiring large businesses to make "a good-faith effort to connect a customer to a human within 15 minutes," to tell customers when the representative is a chatbot, and to list customer-service contact information on their websites. Two days later, he signed the workplace bills; KQED's roundup of the Sept. 30 signings lists SB 947, SB 951, AB 1883, AB 1979 (which keeps AI from replacing licensed clinicians' judgment) and AB 1331 (no surveillance tech in workplace bathrooms), and notes that two other AI bills, AB 2656 and AB 2575, were vetoed.

If you only remember one thing: these laws don't ban AI. They ban letting AI act alone in moments where a person is owed a person.

SB 947: AI can help in HR decisions but can't make them alone

This is the law most small and mid-sized businesses will actually have to comply with.

What counts as an "automated decision system"

The definition is broad. According to Paul Hastings' client alert on the new employer laws, an ADS is "any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence" that affects employment decisions. That reaches well past ChatGPT. Think about:

• A scheduling app that flags "unreliable" employees by attendance score
• A POS or field-service platform that ranks techs or servers by productivity
• Call-center software that scores agents on sentiment or handle time
• A spreadsheet model a manager built to decide who gets written up
• An AI summary of performance reviews that a manager pastes into a termination memo

If any of those drives a write-up or a firing, SB 947 applies.

The three rules

1. Never "solely." You can't discipline or terminate someone based only on what the system says.
2. If "primarily," a human corroborates. When an ADS is the main basis, HR Dive reports that "a human must corroborate the decision using supporting information such as gathered evaluations, personnel files and other documents."
3. Tell the worker, in writing. The employee gets written notice that the employer relied primarily on an ADS and that a human corroborated the output. Workers can request a description of the data the system used (Paul Hastings notes it can be in anonymized form), and the human reviewer needs to be able to explain the decision.

Penalties and enforcement

Paul Hastings puts the civil penalty at $500 per violation, enforced by the Labor Commissioner and public prosecutors, with no explicit private right of action for employees. $500 sounds small. It isn't the main risk. The bigger exposure is that a missing notice or a rubber-stamp review becomes Exhibit A in a wrongful-termination or discrimination claim you would have faced anyway.

Related rules already on the calendar

SB 947 isn't the only employment-AI rule arriving. HR Dive also points out that California Privacy Protection Agency regulations, effective Jan. 1, 2027, require businesses covered by the CCPA that use AI for hiring or compensation decisions to give pre-use notice, offer opt-outs and complete risk assessments. And if you have 75 or more employees, SB 951 adds AI-specific disclosure to layoff notices starting the same day, including the type of AI involved and the statement "This notice is for a technology displacement," per Paul Hastings.

AB 1609: the 15-minute human handoff and chatbot disclosure

AB 1609 targets the experience everyone hates: a chatbot that loops forever and never lets you reach a person.

What it requires of covered businesses

Based on the governor's release and the San Francisco Standard's Oct. 15 report:

• The 15-minute clock starts when the customer asks for a human. As the Standard puts it, the window "applies from the point you basically say, 'Can I speak or chat with a human being?'"
• If you can't connect in time, schedule a callback or appointment within one business day.
• Disclose the bot. Customers must be told when they're dealing with a chatbot, and it can't be passed off as a person.
• Publish contact info on your website so customers know how to reach service.
• Penalties of up to $5,000 for a first violation and $10,000 for later ones, per the Standard.

Companies without phone support don't have to set up a call center, as long as they offer other ways for a customer to "promptly connect with a customer service agent." The Standard's example is pointed: the law reaches AI companies with chatbot-first support, Anthropic included, and the escalation requirement extends to the outsourced chatbot vendor running it.

Why a small business should care anyway

The $500 million threshold leaves out nearly every clinic, law office and restaurant group in Southern California. But three things pull you in anyway.

1. You're already covered by the BOT Act. California has required bot disclosure since 2019. Business and Professions Code section 17941 makes it unlawful to use a bot online to mislead someone about its artificial identity in order to drive a purchase or sale, and it protects you if the disclosure is "clear, conspicuous, and reasonably designed to inform" people that they're talking to a bot. Any AI chat widget that books paid consultations or sells services sits near that line. A one-line disclosure costs nothing.

2. Your customers will expect it. Once the biggest companies your customers deal with have to offer a human within 15 minutes, a small business whose AI receptionist traps people looks worse by comparison. SurveyMonkey's customer service statistics found 79% of Americans strongly prefer a human over an AI agent and 89% believe companies should always offer the option to speak with a human.

3. Your vendors are covered. If you run on a large platform's booking, payments or support tools, their escalation flows will change, and your own setup should match.

"We're too small for this law": why the standard still reaches you

Here's the honest scorecard for a typical SoCal small business:

The pattern: SB 947 is the law you must comply with; AB 1609 is the bar you should design to. Getting your chatbot right now avoids rebuilding it if the threshold drops in a future session, and it costs about the same as getting it wrong.

Designing AI receptionists and chatbots with a built-in human exit

This is where most of the practical work lives, and where we spend a lot of our AI automation builds. A good human exit is a design choice made at the start, not a patch.

1. Disclose in the first line

Open with something like: "Hi, I'm the AI assistant for Coastal Family Dental. I can book, reschedule or answer questions, and I can get you to a person anytime. Just say 'human.'" That one sentence covers BOT Act disclosure, matches AB 1609's spirit and reduces frustration.

2. Make "human" a hard trigger

Treat phrases like "agent," "person," "representative," "talk to someone," or repeated "no" as an instant escalation, not something the model weighs. The clock in AB 1609 starts at the request; your design should too.

3. Have a real place to send them

The handoff has to land somewhere:

• During business hours: warm transfer to the front desk, or a text to the on-duty person with the conversation summary attached.
• After hours: offer a callback slot within the next business day, booked directly into your calendar, and confirm it by SMS.
• Urgent keywords (medical emergency, gas smell, flooding, legal deadline): bypass the bot entirely with clear instructions.

4. Pass context, not just the caller

Nobody wants to repeat themselves. The bot should hand the human a three-line summary: who, what they want, what's already been tried. This is mostly a context engineering problem: deciding what the model keeps and what it passes along.

5. Log every escalation

Record when a human was requested, when one responded, and the outcome. If you ever need to show good faith, or just want to know whether your front desk is overwhelmed, the log is the evidence.

A worked example (hypothetical)

Take a hypothetical 25-person physical therapy practice in Pasadena running an AI receptionist on its main line and website. Before the changes, the bot answered everything, kept trying to reschedule when patients asked for a person, and its "manager review" of no-show patterns quietly fed into staff write-ups for front-desk errors.

After a redesign:

• The bot introduces itself as AI, and "person" or "human" routes to the front desk within the same call during hours, or books a next-business-day callback after hours.
• Escalations post to a shared channel with a summary; median response time shows up on a weekly report.
• The no-show dashboard still exists, but any write-up it prompts requires the office manager to attach the schedule records and a note from the employee conversation, and the employee gets a written notice template explaining that a system flagged the issue and a person reviewed it.

None of that requires new software. It needs a few rules written down and wired into tools the practice already pays for.

A compliance checklist before July 2027

Work through this in order. Most businesses can get through the first five items in a single afternoon.

Customer-facing AI (chatbots, AI receptionists, SMS bots)

• [ ] List every place a customer can talk to an automated system: website chat, phone tree, AI receptionist, text, social DMs.
• [ ] Add a clear "I'm an AI assistant" disclosure to the first message of each.
• [ ] Make "human/agent/person" an instant escalation trigger.
• [ ] Define where escalations go during hours and after hours, with a next-business-day callback option.
• [ ] Publish real contact information (phone or email and hours) on your website.
• [ ] Start logging escalation requests and response times.

Employee-facing AI (SB 947, SB 951, AB 1883)

• [ ] Inventory every tool that scores, ranks, flags or predicts employee behavior, including features buried inside scheduling, POS and field-service software.
• [ ] Mark which ones ever inform discipline or termination.
• [ ] Write a one-page policy: no discipline or firing based solely on a system output; any ADS-driven decision requires a named reviewer who checks evaluations, personnel files or documented incidents and writes down their reasoning.
• [ ] Draft a written notice template for affected employees and a process to answer requests for a description of the data used.
• [ ] Train managers so they can explain a decision without saying "the system said so."
• [ ] If you have 75+ employees, add AI-displacement language to your layoff-notice process.
• [ ] Confirm none of your monitoring tools infer emotions or collect neural data (rare in small business, but check vendor feature lists).
• [ ] If you're covered by the CCPA, map hiring and pay tools against the CPPA's ADMT rules, effective Jan. 1, 2027.

Vendors

• [ ] Ask each AI vendor in writing: does the product make or recommend employment decisions, can you export the data it used, and can the bot hand off to a human on request?

What we'd do

If we were sitting with a Southern California owner today, we'd do three things in order:

1. Map the automation you already have. Most owners undercount it. The scoring feature inside the scheduling app is the one that causes SB 947 problems.
2. Fix the customer exit first. It's the fastest win and the most visible one. Disclosure plus a reliable human handoff improves customer experience the same week.
3. Put the human-review step into the tool, not a memo. If the dashboard can't produce a write-up without an attached reviewer note, the policy enforces itself.

That's the core of our AI consultation for California businesses: an inventory of where AI touches customers and employees, a ranked list of fixes, and a plan that fits the tools you already pay for. We work with clinics, law and accounting offices, trades and restaurant groups across Southern California.

Next step

July 2027 sounds far off, but the inventory, policy and chatbot redesign take a few weeks of real attention, and it's easier before your next software renewal than after. Book a free automation audit and we'll show you where AI is already making decisions in your business, and how to keep a person in the loop for every one that matters.

Sources

1. Governor Newsom signs commonsense legislation to make your life easier — Office of the Governor of California, September 2026 — AB 1609 signed: 15-minute good-faith human connection or appointment, chatbot disclosure, website contact info.
2. Newsom Signs Slate of AI Workplace Laws, Barring 'Robo Bosses' and Surveillance — KQED, October 2026 — SB 947, SB 951, AB 1883, AB 1979 and AB 1331 signed Sept. 30; SB 947 effective July 1, 2027; AB 2656 and AB 2575 vetoed.
3. California revamps AI protections for workers in flurry of bill signings — HR Dive, October 2026 — SB 947 human corroboration, written notice and data-description rights; CPPA ADMT rules from Jan. 1, 2027.
4. Gov. Gavin Newsom Signs New AI Laws Affecting California Employers — Paul Hastings, October 2026 — SB 947 covers all CA employers, broad ADS definition, $500/violation, no explicit private right of action; SB 951 (75+ employees) and AB 1883 from Jan. 1, 2027.
5. Anthropic uses AI to dodge customer complaints. A new law may force a change — The San Francisco Standard, October 2026 — AB 1609's clock starts at the request for a human; callback within one business day; $5,000/$10,000 penalties; applies to chatbot-first AI firms.
6. Customer service statistics — SurveyMonkey, February 2026 — 79% strongly prefer a human over an AI agent; 89% want a human option always available.
7. California Business and Professions Code section 17941 — California Legislature (BOT Act), operative July 2019 — bots may not mislead people about their artificial identity to drive a sale; clear, conspicuous disclosure avoids liability.