The Pope’s AI Guy Is Worried About ‘Cartel’ Behavior Among Big Labs

Published: 2026-09-24
One wide pipe feeds a glowing funnel while many smaller pipes sit dry and cracked nearby.
When a few labs control the pipeline, everyone downstream drinks from the same narrow spout. AI-generated illustration

What Did the Pope's AI Adviser Actually Warn About?

Paolo Benanti, a Franciscan friar and ethics professor who advises Pope Francis on artificial intelligence, has warned that a small group of dominant AI companies risks behaving like a cartel — concentrating control over the models that much of the world now depends on. That is the core of the warning. The concern is about market structure and power, not about any specific product or price.

Here is the honest limitation up front: I cannot give you a verbatim quote, the exact venue, or the precise date of Benanti's remarks. I have not verified the original interview or transcript, and this article's reference material does not contain them. If you need the primary source, find the original interview — do not take my paraphrase as a citation. What I can do is explain what the concern means in practice, because the structural problem he is pointing at is real and it affects what you can do with AI tools right now.

Why 'Cartel Behavior' Is the Right Frame — and Where It Breaks Down

A cartel, in the economic sense, is a group of competitors who coordinate to control supply, price, or access. Benanti's worry is not that executives are meeting in smoke-filled rooms. It is that the economics of frontier AI naturally push toward a handful of players: training runs cost enormous amounts of capital, compute is scarce, and the best models create a feedback loop where more users mean more data and more revenue to fund the next model.

That is a structural argument, and it is defensible. Where it breaks down is enforcement. Calling something cartel-like is a description of an outcome, not proof of coordination. Two companies charging similar prices is not evidence they agreed to — it can simply mean they face the same costs. Anyone who tells you parallel pricing proves collusion is skipping a step.

The Concentration Test: A Decision Rule You Can Actually Use

A scale tips down on the side holding a tight cluster of identical spheres, not the scattered shapes.
The concentration test weighs how few suppliers hold how much of your stack — not how good their demos look. AI-generated illustration

Here is a practical way to apply Benanti's concern to your own tooling decisions. Ask three questions about any AI tool you depend on:

None of this is AI-specific in the abstract — vendor lock-in is an old problem. What makes it sharper here is the pace. Model capabilities shift on a scale of months, so a tool that was the obvious choice last quarter can be behind this quarter, and your migration cost does not shrink just because the market moved.

A Worked Example: Choosing an Image Tool Under Concentration Risk

Say you run a small design shop and need AI image generation. Two obvious options are Midjourney and Adobe Photoshop. Their entry points differ: Midjourney's Basic plan is $10 per month, while Adobe's Photography Plan — which bundles Lightroom and Photoshop — is $9.99 per month, and Photoshop on its own is $20.99 per month. Adobe's All Apps tier runs $54.99 per month.

Notice what that comparison does and does not tell you. It tells you the entry prices are close. It tells you nothing about coordination — these are different companies with different cost structures selling different things, and similar round numbers are not evidence of anything. Treating two similar prices as a symptom of cartel behavior would be exactly the kind of leap Benanti's warning does not license.

What the comparison is genuinely useful for is the substitutability question. If your workflow is built on Photoshop's layer model and file formats, switching to Midjourney is not a switch — it is a different job. That is the real concentration risk, and it lives in your workflow, not in the price list.

What AI Does Badly in This Scenario

If you try to use an AI assistant to research this topic, be careful. Language models will happily produce a confident-sounding quote attributed to a named public figure, complete with a plausible date and publication, that does not exist. This is the single most common failure mode when you ask an AI about recent news involving real people.

The fix is procedural, not technical: for any claim about what a named person said, require a link to the primary source before you act on it. If the model cannot produce one you can open, treat the claim as unverified. That rule would have caught the gap in this very article — I am telling you plainly that I could not verify Benanti's exact words, rather than inventing them.

The concentration risk in AI is not that companies secretly agree on prices. It is that your workflow quietly becomes unportable, and you find out only when the terms change.

Where This Advice Fails

A wall of identical keyholes with a single key hanging in front, suggesting one gatekeeper.
If every door needs the same key, switching tools is not a choice — it is permission. AI-generated illustration

Two honest caveats. First, diversification has a cost. Running two image tools, or maintaining a fallback model, means paying twice and maintaining two sets of prompts. For a solo operator, that overhead may exceed the risk. The decision rule above is a prompt for thinking, not a mandate.

Second, the market is genuinely moving. Pricing and plan structures change frequently — the figures above are a snapshot, and the vendor's own page is the only reliable source for what you will actually pay today. Any analysis built on a price list has a shelf life measured in months.

For teams that generate a lot of content, one part of the workflow worth examining is prompt overhead itself. Tools like AI-Mind take a described goal and a content type and handle the prompt construction, which reduces how much of your process is tied to one vendor's specific prompting quirks — a small but real reduction in switching cost.

Key Takeaways

The Practical Move

You do not need to resolve whether big AI labs are behaving like a cartel. You need to know how long it would take you to leave any tool you depend on. Pick your most important AI tool this week and answer that question honestly. If the answer is uncomfortable, that is the finding — and it is more actionable than any headline about what a Vatican adviser did or did not say.

For related reading on keeping control of your AI workflow, see How to Use AI With Your Privacy Intact and Your RAG Finds the Documents. But Which Ones Should Reach the LLM?

Sources

Frequently Asked Questions

Who is the Pope's AI adviser and what did he warn about?

Paolo Benanti is a Franciscan friar and ethics professor who advises Pope Francis on artificial intelligence. He has warned that a small group of dominant AI companies risks behaving like a cartel, concentrating control over the models much of the world depends on. The concern is about market structure and power rather than any specific pricing agreement. I could not verify his exact words or the venue, so check the original interview if you need a direct quote.

Does similar pricing between AI companies prove they are coordinating?

No. Two vendors charging similar amounts is not evidence of an agreement — it can simply mean they face similar costs for compute, talent, and infrastructure. Proving coordination requires evidence of actual communication or agreement, not parallel price points. Treating similar prices as proof of collusion skips that step entirely. The useful question is not whether prices match, but how easily you could leave either vendor.

Related: I've explored this before in best ai book writing assistant.

How do I reduce my risk from AI market concentration?

Focus on switching cost rather than price. Ask how many days it would take to move your workflow elsewhere, how much of your data and history lives with one vendor, and whether the tool routes across multiple models or depends on a single upstream provider. Diversifying carries real overhead, so weigh it against the exposure you actually have. A tool you could leave in a week is a very different risk than one you could not leave at all.

Related: This connects to what I wrote about If the AI Industry Followed Its Own Research, It Might Ha....

How this article was produced: it was generated by an automated content pipeline from the sources listed above. No human editor wrote or reviewed it, and we did not personally test the tools described. Facts and prices that appear here come from our own AI tool database, and its verification date is noted where relevant. Spotted an error? Tell us and we will correct or remove it.

Want to try this yourself? AI-Mind generates content from a plain description — no prompt engineering required.

Try AI-Mind