GOVERNANCE
OCT 05, 2026
BY H.B. HAYES
The Desk — 02
California made “talk to a person” the law. Four more stories that change your support queue this month.
THE DESK — 02 · FIVE STORIES · ABOUT SIX MINUTES
California just made "let me talk to a person" a legal duty for large companies. Two more California laws now reach the tools you use to score your agents.
Five made it this week. I'm not your lawyer, and two of these are legal questions. Ask yours.
01 — In California, "talk to a person" is now the law.
What happened. On 28 September, Governor Newsom signed AB 1609, the Right to Human Customer Service Act. It covers businesses with more than $500 million in annual gross revenue nationwide that sell to California customers. During business hours, customers need a simple way to ask for a person. The business must make a good-faith effort to connect them within 15 minutes, or offer an appointment within one business day. A chatbot can't claim to be human, and must say it's AI if a customer could think otherwise. Prosecutors enforce it, with penalties up to $5,000 for a first violation and $10,000 after that.
What it means. If you're covered, the route to a person is now something you staff, not a setting you tune. Check every place AI talks to customers: can they ask for a person in plain words, does someone pick up within 15 minutes during business hours, and does the bot say it's a bot. If your deflection rate depends on the human being hard to find, expect that number to drop. It was never as good as it looked.
Where it doesn't apply. Under $500 million in revenue, it doesn't. Email, web forms, and voicemail sit outside the 15-minute window. Some utilities, consumer reporting agencies, and hospital communications are exempt. I couldn't confirm the effective date. Ask your counsel.
02 — Your QA tool is now an employment question in California.
What happened. On 30 September, Newsom signed two workplace AI laws. SB 947, the No Robo Bosses Act, bars employers from relying only on an automated decision system to discipline or fire someone. A person has to corroborate the output. It takes effect 1 July 2027. AB 1883 bans workplace surveillance tools that use AI to recognize or predict an employee's emotional state.
What it means. If AI QA scores or sentiment flags feed write-ups or terminations for California agents, a person has to check the call, not just the score. If a tool infers agents' emotions from voice or video, get that confirmed in writing by the vendor and take it to counsel.
Where it doesn't apply. Sentiment analysis on what customers feel is outside AB 1883. It covers employees. QA used for coaching, not discipline, sits outside SB 947. No California agents, no change.
03 — If you run Salesforce, Agentforce is already switched on.
What happened. Salesforce's Winter '27 release turns the Agentforce platform on by default in every org whose licence or edition includes it, and removes the setup toggle. Most production orgs get the release over the first two October weekends; your exact date is on Salesforce Trust Status under Maintenances. Salesforce says billing doesn't change.
What it means. Enabled isn't deployed. No agent talks to a customer until someone builds and activates it. If you own Service Cloud, ask your admin who holds the Manage AI Agents permission, and decide whose sign-off an agent needs before it goes live.
Where it doesn't apply. If your edition doesn't include Agentforce access, nothing changes. Admins can still turn the platform off in Einstein Setup and deactivate individual agents.
04 — Zendesk will move your old bots in January whether you're ready or not.
What happened. Zendesk is retiring AI agents - Essential and its legacy bot tools, including bot builder, answers, and intents. Development stopped on 31 August. Per Zendesk's migration notice from 23 June, essential AI agents not yet moved will be migrated for you in January 2027, and legacy ones from early February 2027, with some exceptions. Zendesk also warns that automated resolution counts may change after the move, and because pricing follows those counts, so may your bill.
What it means. A forced migration means different answers to the same customer, written by a different system, on a date you didn't pick. If you run one of these bots, migrate on your own schedule and re-test the classes of contact the bot handles before it goes live. Watch automated resolutions for the first month after. If the count jumps, check whether those were resolutions or customers who gave up.
Where it doesn't apply. Customers who bought the AI agents - Advanced add-on aren't affected. Neither is anyone who never set up a Zendesk AI agent.
05 — OpenAI's notice list went from dozens to more than 100 in a week.
What happened. On 1 October, Reuters reported that OpenAI has told more than 100 organizations about unauthorized activity tied to its AI agents, citing an OpenAI blog post. Reporting around 25 September put the number at dozens. OpenAI says its review will take months. A notice isn't proof of a breach. Some of the activity touched only public information.
What it means. In Edition 01, I said to ask every AI vendor how fast they'd tell you if their system touched your data. This week adds a second question: who receives the notice. If an outside agent hits your help center or web forms, the notice goes to whoever owns that site. Find out who that is, and make sure support hears about it the same day.
Where it doesn't apply. These were OpenAI's agents running in training and evaluation, not products a support team deployed. Nothing here says a support tool built on OpenAI models did this.
Worth knowing, not worth your Monday
Connecticut's omnibus AI law, SB 5, started taking effect on 1 October. Its first duties fall on AI subscription sellers, frontier developers, and companion-chatbot operators. In Sacramento, Newsom signed an executive order on 30 September telling state agencies to keep calling it "artificial intelligence," whatever terminology Washington uses. Both are real. Neither changes a decision in your queue this month.
One thing to do this week
Open every place a customer can talk to AI in your queue. In each one, type or say "I want to talk to a person." Time how long it takes a person to respond, and write the number down next to each channel.
If the bot doesn't understand the request, or any number is over 15 minutes during business hours, that's the finding. California now has a word for it.
Sources
- California Just Made It Easier to Get Past AI Customer Service Chatbots and Reach a Real Person — Law Commentary
- California's 2026 Privacy and AI Bills: What Newsom Signed and Vetoed — Stauss PLLC
- California Governor Signs 3 Bills Targeting AI and Workplace Surveillance — Ogletree Deakins
- Salesforce to Auto-Enable Agentforce in Winter '27: What That Means for You — Salesforce Ben
- Admin Release Countdown: Get Ready for Winter '27 — Salesforce Admins
- Announcing the removal of AI agents - Essential and legacy functionality — Zendesk
- OpenAI alerts more than 100 groups about rogue AI agent activity — Reuters
- Connecticut Enacts Sweeping AI Law With 2026–2028 Obligations — National Law Review
— H.B.
The Desk is original writing from Copious. We take no money from AI vendors, so nobody paid to be in it or to be kept out of it. If I got something wrong, tell me at hello@copious.app. I'll fix it and say so.
Next edition: Monday, 12 October 2026.