AI Radio is an AI-generated audio program. The hosts and callers are synthetic voices, not real people, and the figures and claims are as stated in the program. Check them against primary sources before relying on them. This is general information, not legal, insurance, or compliance advice.
Take the next step
Each week's AI Radio follows the same three questions as the written posts: who decides, what is covered, and where the file is. Start with the free General Business Starter Packet or, if you run a senior housing organization, the senior housing packet.
Read the transcript
0:00 Paul: Welcome to AI Talk Radio. I’m Paul, broadcasting from our London studio. Today, we’re looking at a pivotal shift in the deployment of artificial intelligence across the enterprise landscape, specifically in the sensitive sectors of senior housing and insurance. The overarching theme this week is clear: the machine can advise, but someone still has to decide. We’re joined by a roundtable of listeners to dive into the technical and governance hurdles of these new frameworks. Sarah in San Francisco, Mark in Chicago, and David in New Jersey. Welcome to the show.
0:39 Sarah: Hey Paul. Yeah, it’s uh... it’s been a massive week here in California. The Governor just signed off on a whole suite of bills that really, you know, draw a line in the sand about where the software ends and the professional begins.
0:54 Paul: Let’s start there, Sarah. You’re looking closely at the healthcare bills, AB 1979 and SB 503. What’s the core technical friction there?
1:07 Sarah: Well, it’s really about the... uh... the clinical scope of practice. So, AB 1979 basically says that if you’re a health facility, you have to ensure a licensed professional keeps their independent judgment. You can’t let the AI, like, direct or supervise unlicensed staff—like Certified Nursing Assistants, or CNAs. Imagine a night shift, right? A CNA has a tablet, and the AI flags a fall risk for a resident. The law now says the AI can’t just tell the CNA what to do next if that action is something only a licensed nurse is legally allowed to decide. It can inform the nurse, but it can’t... it can’t bypass them to instruct the aide.
1:52 Mark: And that’s a huge deal for the, uh, the ROI models these facilities were building, right?
2:00 Paul: Mark in Chicago, you’re looking at the risk side of this.
2:02 Mark: Exactly. I mean, thanks Paul. See, administrators have been buying these systems specifically to, you know, bridge the staffing gap. If the Director of Nursing has to sign off on every single AI-generated prompt now... well, that’s a bottleneck. But SB 503 adds another layer. It says if you deploy a tool with a "foreseeable risk of biased impact," you’re legally responsible for mitigating it. It’s not just on the vendor anymore. The operator is now... they're in the hot seat for the math.
2:33 David: It’s funny you say that, Mark, because I’m seeing the same thing on the IT side. I’m David, by the way, calling from New Jersey. We’ve been watching the Bridgeway pilot over here in Hillsborough. They’re using this European tech called "voize"—it’s like a voice-documentation AI. The caregivers dictate their observations and vitals during the shift. But the key thing—and this ties into what Sarah was saying—is the staff have to *verify* the structured notes before they go into the record. The AI does the heavy lifting of formatting, but the human is the gatekeeper.
3:09 Paul: That verification step seems to be the "Decision Owner" framework in action. David, stay with that for a moment. How does that impact the actual efficiency? Does it defeat the purpose of the automation?
3:18 David: Actually, no. I mean, the data coming out—there was this Charité study in the Journal of Medical Internet Research—it showed a 27% reduction in documentation time. So even with the human "double-check," you’re still gaining hours. It’s about, uh... it’s about moving the human from a "data entry clerk" to a "data validator."
3:39 Sarah: Right, and that’s what the "No Robo Bosses" law, SB 947, is trying to do for management, too. You can’t just have a system flag a worker for low productivity and, you know, automatically fire them. If the AI is the primary basis for discipline, a human manager has to corroborate it with "independent information." Like, you actually have to look at the peer reviews or talk to the person. You can't just... click "delete" on a human career because the algorithm said so.
4:09 Mark: But Paul, here’s where it gets really interesting for the tech-savvy crowd. The insurance market is finally hitting back. For most of 2026, we saw these insurance carriers just dropping "AI exclusions" into policies. Like, "Oh, you used AI? We’re not covering that failure." But now the big brokers—Marsh, Aon, WTW—they’re drawing a line. Nick Studer from Marsh basically said they won’t support broad AI exclusions that leave gaps for clients. They’re telling clients to literally find other markets if a carrier insists on a blanket exclusion.
4:46 Paul: That’s a significant shift in leverage, isn’t it? The brokers are essentially calling the exclusions a "commercial disaster."
4:55 Mark: It is! But it’s nuanced. Andrew Hill at Willis Towers Watson—WTW—he’s warning that while we shouldn’t have blanket exclusions, AI *is* a "systemic exposure." It’s not like a fire in one building. If a common LLM or a specific clinical tool fails, it fails across ten thousand facilities at once. That’s what’s keeping the reinsurers awake.
5:21 David: So Mark, does that mean companies need, like, a specific "AI policy" now? Instead of just general liability?
5:30 Mark: Yeah, that’s where the Gallagher Re research is pointing. They did this big report with MIT recently. Most vendor contracts have these tiny liability caps. If you’re a senior living home and the AI messes up a medication dosage, the vendor might only be liable for, say, the cost of the software subscription. That leaves the facility exposed for millions. So the recommendation now is a 30-day audit. You’ve got to ask your broker: "Does this policy respond to an AI failure?" and "Where is the indemnification in the vendor contract?"
6:11 Sarah: It’s really about this "Decision Owner" concept, isn't it? Like, who is the one person at the company who actually *understands* what the tool is doing? I was reading about Solera Senior Living. They’ve got this "agentic AI" reviewer. It’s basically an AI that audits their other systems.
6:28 Paul: An AI auditing the AI? Tell us more about that, Sarah.
6:33 Sarah: So, they have ambient sensors for fall detection. If the sensor sees a fall, but no human wrote a report, the AI flags a "charting gap" to the care team. But—and this is the cool part—if a nurse *does* report a fall but the sensors missed it, the AI routes that to the tech vendor as a "detection failure." They call it bi-directional auditing. The results were wild... they went from reconciling one-in-eight falls to six-in-seven.
7:05 David: See, that’s a smart use of data you already have. There’s another one—Longevity Senior Living. Their president, Carl Hirschman, had this great quote. He said you don’t need more data; you need "signal through the noise." They built a clinical tracker that looks at everything—missed meds, UTIs, absences—and it just flags the uncompleted tasks. In 90 days, their medication issues dropped by 78%. And the clinical directors saved like two hours a day because they weren't digging through charts manually.
7:42 Paul: It sounds like the successful firms are those who treat AI as a workflow tool rather than a replacement for professional oversight.
7:49 Sarah: Exactly. But you have to be so careful with the "who." If I’m an operator, I’m looking at every tool by January 1st. I’m asking: Does this tool talk to an unlicensed aide? If yes, I need the Director of Nursing to sign off on the protocol. If I don’t have a named "Decision Owner" for that algorithm, I’m basically just waiting for a regulatory audit or a lawsuit.
8:14 Mark: It’s funny, we used to talk about "shadow IT." Now we’re talking about "shadow AI." Managers just... uh... plugging in tools to make their lives easier without checking the insurance implications or the California Labor Code.
8:31 Paul: And the "No Robo Bosses" act, as you mentioned Sarah, really puts the brakes on that. If you're using AI to monitor productivity—even things like predicting a worker's emotional state—California is saying "no." They’ve even banned tools that collect neural data or surveillance in bathrooms. It’s a very human-centric approach to a very technical problem.
8:54 David: It’s going to be a patchwork for a while, though. If you’re a national chain, do you have one set of rules for your California homes and another for, say, New Jersey? Most of the experts are saying... uh... just adopt the highest standard. Make "human-in-the-loop" your national policy so you don’t have to keep track of forty different state laws.
9:17 Paul: It’s a fascinating moment where the legal framework is finally catching up to the capabilities of the hardware. We are moving from the "move fast and break things" era of AI to the "audit often and assign ownership" era.
9:31 Sarah: Yeah. It’s about accountability. If the machine makes a recommendation, that’s advice. If a human clicks "approve," that’s a decision. And the law is making sure we know exactly whose finger was on that button.
9:47 Paul: A vital distinction. Thank you, Sarah, Mark, and David for your insights today. To wrap up, the message for leadership in these tough times is clear: inventory your tools, identify your Decision Owners, and ensure that your insurance coverage matches the reality of your digital stack. The machine can advise, but the responsibility remains, as ever, human. This is AI Talk Radio. Thank you for listening.