Intuit (NASDAQ: INTU) was supposed to be safe. Switching tax software is a chore nobody wants. Then in June, a Goldman Sachs analyst estimated that AI can prepare a standard tax return for about $0.12, compared with up to $162 for TurboTax. At its low point in late June, Intuit stock was down about 61% from its peak.
Rollins (NYSE: ROL) was supposed to be even safer. No model can kill termites. But Rollins missed its second-quarter numbers because fewer homeowners found Orkin online, and analysts have pointed to AI search as a possible cause.
Meanwhile Fair Isaac (NYSE: FICO), the credit score company everyone called a monopoly, lost its exclusive mortgage pricing on Sept. 29. That one was regulators, not AI.
So most “AI-proof stock” lists ask the wrong question. Whether AI can do the job matters less than whether AI changes who controls the customer or the price.
I screened 18 companies against that question. Five held up. Each one has a moat built from something a model can’t copy: a regulatory license, a network, data nobody else has, or physical land and permits.
What AI actually does to a business
AI makes certain work cheap: answering questions, writing code, explaining complicated rules. A company that charged for making complexity easy now has a pricing problem. That’s Intuit.
There’s a quieter mechanism too. AI is becoming the front door. When people ask a chatbot instead of running a search, the company that used to buy that click can lose its place in line. That’s Rollins.
The moats that survive both pressures share one trait: they’re expensive to rebuild even with unlimited software. A credit rating needs regulatory recognition. A payment network needs billions of cards and millions of merchants. A landfill needs a permit that can take a decade. Better code doesn’t shortcut any of that.
My test: could a well-funded startup with the best model available rebuild this in five years? For these five, I don’t think so.
1. S&P Global (NYSE: SPGI)
S&P rates corporate and government debt and owns the S&P 500 along with thousands of other indexes. Bond issuers pay it for ratings. Fund companies pay it to use index names.
A model can analyze a balance sheet. It can’t issue a rating that regulators and bond buyers recognize, and no AI lab can launch a fund that tracks the S&P 500 without paying S&P.
Competitive Moat. Ratings revenue rose 17% to a record in the second quarter, and the index business posted its 13th straight record quarter, per S&P’s Q2 2026 earnings release. ETF assets tied to its indexes ended June at $6.35 trillion.
Management Quality. S&P spun off its Mobility division on July 1 and raised its 2026 buyback target to more than $7 billion. It’s also using AI on itself, with nearly 60% of a targeted $100 million in annual cost savings already captured, per the company’s Q2 call.
The risk: Market Intelligence, the data and research division, is the piece most exposed to AI, since customers can find some of that analysis cheaper elsewhere. Ratings revenue also moves with debt issuance, so a slow year for bond deals would show up fast.
Valuation: At $408.57 on Oct. 14, the stock trades around 25 times trailing earnings. At each year-end since 2021, that multiple ranged from roughly 33x to 53x, based on reported annual earnings. The stock sits about 26% below its 52-week high.
2. Verisk (NASDAQ: VRSK)
Verisk collects policy and claims data from across the U.S. property and casualty insurance industry and sells analytics built on it back to insurers.
AI models are only as good as their data. Verisk holds decades of pooled industry history that insurers contributed, which makes it more likely to supply AI than be replaced by it.
Competitive Moat. Insurers keep contributing because Verisk’s standards are built into how the industry prices and files coverage.
Business Quality. Underlying subscription revenue grew 8% in Q2 on an organic constant-currency basis, ahead of total organic growth of 5.8%, per Verisk’s Q2 2026 release.
The risk: This is the name where management gives me the most pause. Verisk agreed in 2025 to buy roofing software maker AccuLynx for $2.35 billion, then tried to walk away when the FTC review dragged on. In August, a Delaware court ruled the termination invalid because Verisk’s own conduct caused the regulatory delay, and ordered it to keep pursuing the deal. The court found no sign of bad faith, and Verisk has appealed. It’s still a real stumble in capital allocation, which I weigh heavily.
Valuation: At $180.58, Verisk trades near 28 times trailing earnings, below the roughly 29x to 57x range at each year-end since 2021. It’s about 26% below its 52-week high.
3. Mastercard (NYSE: MA)
Mastercard runs the network that moves money between a cardholder’s bank and a merchant’s bank, taking a small fee on each transaction.
If AI agents start shopping on people’s behalf, they still need a trusted way to pay, and Mastercard has already launched an agentic payment capability. More AI also means more sophisticated fraud, and Mastercard sells the security tools banks use to fight it.
Competitive Moat. With about 3.7 billion branded cards worldwide, every new card makes the network more useful to merchants and the reverse. That two-sided loop took decades to build.
Business Quality. Second-quarter net revenue rose 14% to $9.3 billion, and value-added services, which include fraud and security tools, grew 20%, per Mastercard’s Q2 2026 release. Net income was $4.4 billion on that $9.3 billion of revenue.
The risk: Regulators in the U.S. and Europe keep pressing on card fees. The longer-term risk is that AI agents settle payments through bank transfers or stablecoins that skip card networks entirely.
Valuation: At $603.24, Mastercard sits near its 52-week high, around 33 times trailing earnings. That’s at the low end of its 34x to 41x year-end range since 2021, but it’s the most fully priced name here.
4. WM (NYSE: WM)
WM picks up trash and recycling from homes and businesses across North America and owns the landfills where much of it ends up.
You can’t prompt your way to a landfill. New permits are rare, slow, and fought locally. AI mostly helps WM: its recycling automation has cut labor cost per ton by 30%, per the company’s Q2 call.
Competitive Moat. Owning the final destination for waste gives WM pricing power over collection, even over competitors’ trucks that need somewhere to dump.
Management Quality. WM returned $1.04 billion to shareholders in Q2 through $659 million of buybacks and $379 million of dividends, per its Q2 2026 release, while free cash flow grew 35%.
The risk: Volumes are soft. Management expects collection and disposal volumes to stay flat in the second half and trimmed its full-year revenue outlook. This is also a capital-heavy business with leverage near the top of its target range, so higher rates bite.
Valuation: At $215.75, WM trades around 30 times trailing earnings, in the lower half of its 29x to 39x year-end range since 2021 and about 13% below its 52-week high.
5. Copart (NASDAQ: CPRT)
Copart runs online auctions for wrecked and totaled vehicles, mostly for insurance companies. It sold more than 4 million units last year from over 250 locations in 11 countries.
The auction is online. The cars aren’t. Copart’s moat is land: large yards zoned for storing damaged vehicles near major cities, which are slow and expensive to replicate.
Competitive Moat. The yards attract the sellers, and roughly a million registered buyers across more than 185 countries attract the yards’ business.
Financials and Efficiency. Copart carries essentially no debt and ended fiscal 2026 with about $1.91 billion in cash plus $2.58 billion in held-to-maturity securities. Founder Willis Johnson remains chairman and one of its largest individual shareholders.
The risk: Growth has stalled. Fiscal 2026 revenue rose just 0.4%, and fourth-quarter net income fell 17.4%, per Copart’s Sept. 10 release. Insurers are keeping more damaged cars and some drivers are cutting coverage. And this is the one name where AI could hurt the business: driver-assistance systems may reduce accidents over time, which means fewer totaled cars to sell.
Valuation: At $26.27, Copart trades near its 52-week low and about 59% below its high. That’s roughly 17 times trailing earnings, against 28x to 37x at the end of each fiscal year from 2022 through 2025. The price already assumes a lot of that slowdown.
What would change my mind
A regulator opens the gate. FICO showed how fast a protected position can change with one agency decision. Similar changes for credit ratings or index licensing would narrow S&P Global’s moat.
Insurers take their data elsewhere. If carriers start licensing claims data straight to AI model builders, Verisk becomes a middleman.
AI agents pay without cards. If agent-driven shopping settles mostly through bank transfers or stablecoins, Mastercard misses the growth that matters most.
Accidents fall faster than expected. Rapid adoption of driver assistance would shrink Copart’s supply of cars.
The front door moves. If any of these five starts losing customers because an AI assistant sits between it and the buyer, the moat is leaking even if the product is fine.
Margins fall while revenue grows. That’s the clearest early sign that pricing power is slipping.
The bottom line
I’m not predicting which of these stocks will outperform, or when. Some may lag for years.
My point is narrower. The useful question about AI isn’t “is this a software company?” It’s “can a model take the customer or the price away?” Intuit, Rollins, and FICO all looked safe by the old test and still got hit.
These five run on licenses, networks, data, and land, which take years to build, not better code.
Long-term conviction, not short-term noise.