Shelf Life | Vol. 50 — Like a Prayer: The Human Element of AI that Connects Paris Hilton with the Pope
🗓️ May 2026 | ✍️ Shelf Life
AI is finally making real money. Retailers are cutting customer service costs by something north of 30%. CPG teams are running about a quarter of their marketing through generative models. Logistics is pulling weeks out of inventory cycles. The AI council is having a good year on paper.
Also real: everything else. Deepfakes of actual people, including kids. Voice-cloning scams that pulled hundreds of millions out of elderly Americans last year alone. Hiring algorithms quietly screening candidates by zip code. Customer service bots hallucinating contract terms the legal team then had to honor. So the AI council is also having a more complicated year than the slides imply.
The gap between those two stories closes every quarter. The companies that think they only live on one side of it are wrong. The proof tends to arrive on the front page.
Which is why an unlikely coalition has been making noise the last few weeks. Pope Leo XIV gave another AI ethics address. Paris Hilton testified to Congress about deepfakes and online harm. When the Vatican and a Hilton are sounding the same alarm in the same news cycle, it's time to listen. Your AI prioritization model has a missing axis, and it's maybe the most obvious.
Top Shelf Insights
🙏 Pope Leo XIV: AI is being deployed faster than the moral frameworks needed to evaluate it. His worry is that the cost falls on people who never had a vote.
🎤 Paris Hilton on Capitol Hill: deepfakes, voice-cloning scams, AI-generated abuse material hitting real people in real time, and the platforms are losing ground every quarter. The vulnerable pay first.
📊 Most enterprise AI use case prioritization still runs on two axes. Business value. Feasibility. The implications of getting this wrong is catastrophic.
🧭 The 2026 leaders added three more axes: regulatory risk, brand and trust risk, and human impact. Gartner's AI TRiSM research backs axes three and four. The human axis is something less talked about, and equally important.
⚖️ A moral filter for AI use case prioritization is one of the more useful executive projects you can run this year. It pulls regulatory, brand, and human risk into the same workstream instead of three separate ones nobody owns.
The Sermon
Here's what Pope Leo XIV is actually arguing. AI is being deployed at industrial speed without the ethical infrastructure required to keep up. He's specific about where the damage shows up. Medical decisions. Legal proceedings. Hiring. AI replacing human judgment in contexts where dignity is the whole point. AI concentrating economic power so fast that workers lose ground before policy catches up. AI being weaponized to manipulate people who don't realize they're being manipulated, which, if you've ever sat through a recommendation-engine demo, should ring some bells.
His position is plain. Moral guardrails belong before the deployment. They're showing up afterward instead, in the form of regulators with subpoenas and brand teams with crisis comms decks. (You can usually guess the order. Subpoena first. Deck second.)
The EU AI Act, UNESCO, Singapore, even China. They're all heading toward similar requirements from completely different starting points. The Pope is one voice in a much larger coalition.
The Subpoena
Enter Paris Hilton. Reality TV is entering politics and I'm here for it. Popcorn and Loverboy in hand.
She's been testifying about specific harms. Deepfakes of real people, including herself, including children. AI-generated child sexual abuse material proliferating faster than platforms can take it down. The voice-cloning crisis the AARP is begging Congress to legislate.
Her credibility is real. She lobbied Congress on the Stop Institutional Child Abuse Act and won. She's testified at multiple Senate hearings. Senators take her meetings, which surprised some people a few years ago and surprises basically no one now.
Her argument is plain. The tech is outrunning the protections. The people getting hurt are the most vulnerable ones. The platforms shipping the technology haven't built guardrails that match the harm they're enabling.
Pulpit. Press tour. Funny juxtaposition but very serious topic.
The Two-Axis Trap
Most enterprise AI use case prioritization still runs on the same matrix that we've been using to evaluate technology for a decade. Business value on one axis. Feasibility on the other. Pick the top quadrant and your AI strategy is confirmed with a $2m price tag and shiny bow.
That matrix worked when AI was experimental. It doesn't anymore. Three new dimensions matter, and most teams aren't measuring any of them. I sat in a client AI council last month that defined "risk" as whether the model breaks. They were quietly surprised when I suggested risk might also mean the model not breaking.
Regulatory risk. EU AI Act classification, US executive orders, emerging state laws. Different use cases trigger different obligations. Some require pre-deployment audits. Some are quietly disqualifying.
Brand and trust risk. A use case can be technically accurate and still get you a New York Times story if the underlying data was used in a way the customer never agreed to.
Human impact. Whether the use case extends what a person can do or just replaces them. Whether the customer ends up better off after deployment. And whether the value lands with the user, the company, or some honest split between them.
(For the surrounding context: Vol. 25, The Human Machine; Vol. 29, The Hallucination Effect; Vol. 46, AutoCorrect; Vol. 47, Talk to My Agent. The moral filter sits across all four of them.)
The Five-Axis Fix
Here's the framework. Five axes instead of two. Yes, it's more work. And no, you can't farm this one out to AI (or shouldn't).
Business value. Standard.Feasibility. Standard.Regulatory exposure. EU AI Act classification, sector rules, state laws.Brand and trust risk. The worst-case headline if this use case fails publicly.Human impact. Whether it extends capability, preserves agency, and lands real value with the user.
Gartner's AI TRiSM (Trust, Risk and Security Management) research has been pointing at axes three and four for two years. But don't forget axis five plus the scoring model that holds all five together. Everything on one page. A view that lets you see which "high-ROI" use cases actually fail on the human axis before they ship. Soylent green is people, people. Let's do something about it.
On the House
Here's my take.
When the Vatican and Paris Hilton are pointing at the same problem from very different rooms, the niche ethics conversation is over. This is now an executive standard conversation. The companies that operationalize the moral filter first are going to save themselves a lot of regulator meetings. The companies that don't won't have an opportunity to fix it, because they will be reporting to Chief Claude Officer and VP of Gemini.
The strategy gap is upstream of the model choice. The moral filter is where the deck gets re-scored, and where a lot of "obvious" use cases stop looking obvious. I've watched four clients in the last six weeks discover their top "ROI" use case was also their top trust risk.
Three archetypes, three diagnostic priorities.
If you're a Consumer Brand or Retailer:
Score every active AI use case on five axes, not two. The high-ROI candidates are often the high-trust-risk ones in disguise.Stress-test the top three for worst-case public failure. Reverse-engineer the guardrails from the headline you don't want to see. AI is actually great at pointing out risks, and demystifying them. If you're not using Gartner Consulting (achem), then please embrace AI's power to "Red Team" (shoutout to Laura Spillane and Kelsey Warner for teaching me this term during one of our internal AI training sessions, an absolutely amazing Executive-focused AI program designed to upskill Gartner Consultants. Not to belabor the theme here, but way to practice what we preach! Big fan.)Take a public position on AI ethics. Silence reads as evasion. Boilerplate reads as cynicism. Specificity reads like leadership.
If you're a CIO or Chief Data Officer:
Replace the two-axis matrix with the five-axis scorecard. Train the AI council on it before next quarter.Map every active use case against EU AI Act classification and US sector rules.Stand up a review board that includes ethics and brand, not just legal. Most companies have legal. The leaders have all three.
If you're a Board or CEO:
Articulate what AI is actually for in your business. Being ethically responsible. That's hot.Stress-test against three failure scenarios: customer-side hallucination, executive deepfake, biased model in hiring or lending.Decide who in the C-suite owns the moral filter. That's the easiest org gap to fix this year.
This is the work Gartner Consulting is running with executive teams across sectors right now. AI use case prioritization is the first (sometimes second) step in your AI journey, but one of the most important.
By the way, don't get me started on tokenmaxxing and the invisible cost of AI. This will absolutely be a future Shelf Life topic, but also an important part of this matrix.
The Last Look
If our AI strategy doesn't have a moral filter, what does it have? What keeps you up at night about AI and how do we fix it?
It's a real question. And not a theoretical one anymore. Or so, ChatGPT told me.
Jackie Swanson is a Managing Partner at Gartner Consulting, where she advises retailers, fashion brands, manufacturing, utilities and consumer products companies on growth strategy, AI readiness, commerce, and transformation. She lives in New York with her husband and three children, which is either excellent preparation for managing complex client engagements or the other way around. The jury remains out.
📩 Want to talk about what this means for your organization? Book a 1:1 with Jackie → jackie.swanson@gartner.com
Follow Shelf Life on LinkedIn | Substack | Instagram
#ShelfLife #AIStrategy #AIEthics #ResponsibleAI #UseCasePrioritization #AIinRetail #GartnerConsulting