AI Rules for Ads, Health Claims, and Courts
Dear Friend,
In our AI Smart Lab session this week, I brought in something that made me angry. I am in a What’s App channel with about 15 agency owners about in AI marketing, and someone had AI write a quick bio for each of us. The bios for the men said things like “he has already built and exited something substantial” and “we need that person.” Mine said “there is a lot going on here.” Another woman’s said “yes, she belongs in this room too.”
Same run, same tool. The men got credentials. The women got permission to be in the room. The bias is blatant in this case and it is everywhere throughout your outputs.
I asked whether we should train our own model on women’s writing and a matriarchal perspective. Owning a model means owning everything it was trained on, and that is a lot to take on.
Victoria has been working on this issue for herself.. She photographed two years of handwritten journals, page by page, and fed them into her own Claude. It changed how the tool wrote for her. She also keeps a list of writers she trusts, and she asks the model to consider whether those writers would agree or disagree with her argument, not just to copy their tone.
Who is reading this? What should they feel? Is this person being talked down to, or understood? The model checks its work against that before anything leaves.
I wrote a draft of an article on AI as an equalizer for women founders, inside my own account, in a project loaded with our context. The first version came back patronizing. If it does that with all of that context, what is it doing in your business with little or none?
Have you read what your AI writes about your own people? Where do you see the bias in AI show up?
What You Should Get from an AI Consulting Engagement
If you are about to hire an AI partner, you should know what you are owed before you sign anything. Andrea walks through what a real engagement delivers: a roadmap tied to your budget and team size, tool recommendations with the security questions already answered, workflow maps, governance rules, and a first pilot. She is specific about the first 30 days, which should end with a shortlist of 3 to 5 use cases and baseline numbers you can measure against later. She also hands you the checklist for pressure-testing a roadmap, because plenty of them quietly assume people, time, and money you do not have.
Our Most Requested Service: AI Strategy & Implementation
The question clients bring us right now is almost always the same. They know AI can help. They do not know which tools to trust, how to protect their data, or how to get the team to actually use all of it.
We start with your workflows. We map where the time is spent, then build a phased roadmap with secure tools and clear owners, and we train your team to run it after we leave.
One client, a medical training company, worried the whole way through that AI meant cuts. Nobody was laid off. Every person was upskilled and moved into higher-leverage work, and the hands-on training led to automating 99% of their customer service operations, worth $177,000 in combined savings and repurposed labor. We have trained more than 20,217 professionals this way.
This week’s stories
California Now Requires You to Label AI Actors in Ads
Governor Newsom signed Senate Bill 1050 in California on September 16. If a video or audio ad uses an AI-generated performer to sell something, the ad now has to say so clearly. An ad found in violation cannot keep running. SAG AFTRA helped write the bill and hosted the signing.
My Take: The businesses most exposed here are small ones that picked a synthetic presenter because hiring a person cost too much. We have used HeyGen for years for exactly that kind of video, so I am watching this closely. I’ve seen some of these adds that have the AI watermark at the bottom now, wether in anticipation or because it was the free version that they used to create the ad. California is first, which means it becomes the national default inside two years, because most companies don’t build a separate ad for one state.
Does your video ad make clear who, or what, is talking?
Source: Office of the Governor of California
Connecticut Says AI Cannot Deny Your Health Claim Alone
Connecticut’s state comptroller set five AI rules for the carriers running its state employee health plans, covering more than 270,000 members starting January 1, 2027. In the US, this is the first package of its kind at the state plan level. AI cannot be the only thing that denies a claim, and a person has to review it. Member data cannot be used to train other AI models, and carriers have to disclose when AI materially helps decide a benefit. Anthem, Cigna, Aetna, and Caremark are the administrators affected.
My Take: The human review rule will get the headlines, and the training ban is the one that I’m interested in.We use ClickUp partly because it tells us which model is processing our data, which most tools will not. Most platforms that say AI is built in are borrowing somebody else’s model, and you cannot answer a disclosure question about a vendor you cannot see. Connecticut just turned that from a procurement preference into somebody’s legal problem, and I expect other state plans to copy this language rather than write their own.
Need help with your AI policies or protection protocol, we can help, [email protected]
Source: Office of the State Comptroller, Connecticut
Voya Put AI on 7 Million Customer Calls and Trained Everyone First
Voya Financial, a US retirement and benefits company, says AI now handles 7 million customer interactions a year, including 2.8 million calls resolved without a person. Its virtual agents scored 96% positive feedback and resolved 85% of contacts digitally. Nearly all of its global workforce finished an AI literacy certification first, and the company has since reviewed close to 1,000 employee-built AI agents against a governance framework before letting any of them expand.
My Take: Voya certified nearly its whole workforce before it scaled, which is the opposite of how almost every rollout we see happens. Most companies buy the tool, send a link to the vendor’s free training, and wonder why usage flatlines. When we run live trainings with a teams on zoom or in the room, completion rate is around 95%, and the self-serve version rarely gets completed. A financial services company with compliance breathing down its neck still moved faster than the companies treating AI as a software purchase, because it spent the first month on people instead of licenses.
Need help getting your team trained on AI, we can help, [email protected]
Source: Voya Financial
Federal Courts Told Not to Let AI Decide Cases
The Judicial Conference of the United States received its progress report on AI on September 17. The judiciary’s task force has sorted more than 60 separate AI questions into seven working groups. Interim guidance is already out, and it tells US federal courts not to hand core judicial work to AI, including deciding cases or adjudicating them. Everyone working in the courts stays accountable for anything they produce with AI help.
My Take: The US federal judiciary just wrote the AI policy that most businesses still have not. Sixty questions, seven working groups, and one line doing most of the work: the person stays accountable for whatever AI helped produce. We keep what we call a digital suitcase for this, our instructions and training records documented so they can be handed over or moved to another platform, because reconstructing three years of prompts after someone asks for them is not a project anyone wants. Courts move slowly and still got there before most companies did. If the institution with the most to lose from a wrong answer will not let AI make the call, that is a reasonable floor for the rest of us. Have you communicated your stance with your team? Its it documented?
Source: United States Courts
Most of the World Expects AI to Cut Jobs
Pew Research Center surveyed 42,151 people across 36 countries and found that in 34 of 37 places, people expect AI to lead to fewer jobs rather than more. Worry tracks wealth. A median of 55% across 18 high-income countries expect fewer jobs, against 36% across 18 middle-income ones. Australia and South Korea sit at 76%, and the US at 71%, up 7 points since 2024.
My Take: Seventy-one percent of Americans expect AI to cut jobs, which means roughly seven in ten people on your payroll hear every AI announcement you make as being about them. We open engagements with a survey asking what tools people already use, what they use them for, what they wish they could use them for, and what they dislike about AI. It only works when nobody gets punished for the answer. The companies losing this are not the ones automating too fast, they are the ones announcing efficiency gains to a workforce that hears layoffs. Pew ran this fieldwork between February and May, before the latest round of AI job headlines, so the number is probably worse now, not better.
What does your team think you are planning? How are you managing the conversations?
Source: Pew Research Center
Tool picks of the week
- Voice Instructions Creator: Reads your writing to find your tone and style, then builds voice instructions you can hand to any AI tool.
- Open Forge AI: An AI search platform that shows you how your brand shows up in AI answers and where competitors are beating you.
- Leonardo.AI: A free AI image and video generator.
Got a burning question, a fresh take, or just want to share your latest AI wins? Hit us up at [email protected]. Your insights keep this community growing and thriving!
Have a great day!
-Nicole A. Donnelly
Founder, AI Smart Ventures
AI Strategy – AI Training – AI Consulting – AI Implementation
