Pope’s AI Adviser Warns of AI Cartel Behavior
You can feel the AI market tightening around a small circle of companies. That is why the phrase AI cartel behavior matters now, even if it sounds like a charge pulled from an antitrust courtroom. Wired reported that Father Paolo Benanti, a Franciscan friar and AI ethics adviser to Pope Francis, is warning that the biggest labs may be forming a closed club around models, compute, and influence.
His point is not that executives are meeting in a smoky back room. It is subtler, and more useful. If a few firms control the chips, cloud contracts, data pipelines, safety language, and distribution channels, they can shape the market without needing an explicit agreement. For users, startups, publishers, schools, and governments, that can mean fewer choices and weaker accountability.
What Stands Out
- Benanti’s warning focuses on market concentration, not science fiction fears.
- AI power now sits with firms that can afford huge compute bills and scarce Nvidia chips.
- Closed model access can make it hard for regulators, researchers, and rivals to test claims.
- Open-source AI may help, but it does not solve compute concentration by itself.
- Governments need competition policy, procurement rules, and audit rights, not speeches.
Why AI Cartel Behavior Is More Than a Sound Bite
Benanti’s warning, as reported by Wired, lands because the AI race has started to look less like a garage startup boom and more like a capital-intensive infrastructure contest. Training frontier models now requires enormous spending on graphics processors, cloud capacity, talent, and data operations. That filters out most competitors before product quality even enters the conversation.
Look, this is not the same as proving illegal collusion. Antitrust law needs evidence. But cartel-like outcomes can appear when firms follow the same incentives, depend on the same suppliers, and gate access through the same channels. The result can feel coordinated even when no one signs a secret pact.
Benanti’s concern is blunt: if a handful of firms control compute, models, data access, and distribution, competition becomes theater.
Think of it like professional sports. A league can have many teams on paper, but if only three owners can afford elite players, training facilities, and broadcast reach, fans do not get a truly open contest. AI has a similar problem, except the ticket price is compute.
The Pope’s AI Adviser Is Pointing at Power, Not Panic
Paolo Benanti is not a random critic parachuting into the AI debate. He is a Franciscan friar, academic, and ethics specialist who has advised the Vatican and Italy on artificial intelligence. That background matters because the Vatican has framed AI as a human dignity issue, while European policymakers are trying to turn values into enforceable rules.
His intervention also cuts through a lazy split in the AI debate. Too often, people argue over whether AI will save humanity or destroy it. Benanti is asking a more immediate question: who gets to decide how these systems are built, priced, deployed, and monitored?
That should make regulators sit up.
The companies at the center of this discussion include OpenAI, Google DeepMind, Anthropic, Meta, Microsoft, Amazon, and Nvidia. Their roles differ. Some build models, some run cloud platforms, some sell chips, and some do all of the above through partnerships. Still, the shared pattern is clear. AI power is concentrating around firms with rare resources.
How AI Cartel Behavior Could Show Up in Real Life
The danger is not always a price-fixing scheme. It may look like normal business practice. A model provider limits technical documentation. A cloud vendor ties discounts to exclusive use. A lab claims only closed systems are safe. A platform gives its own AI assistant better placement than outside tools.
None of those moves is automatically illegal. Together, they can close the gates. And once customers build on one provider’s APIs, switching becomes costly. You retrain staff, rewrite prompts, adjust compliance reviews, move data, and retest outputs. That friction is a moat.
Watch for these warning signs
- Exclusive cloud or compute deals that lock promising AI firms into one platform.
- Opaque model evaluations where companies publish scores but block independent testing.
- Bundled products that make rival AI tools harder to buy or use.
- Safety claims used as market barriers without clear evidence or public standards.
- Data deals that give dominant firms privileged access to publishers, platforms, or enterprise records.
Here’s the thing. Safety can be real and self-serving at the same time. A lab may be right that advanced systems need controls, while also using that argument to keep outsiders from inspecting its work. Regulators should be allergic to that blend.
What AI Cartel Behavior Means for Users and Startups
For everyday users, concentration often shows up as fewer meaningful choices. Prices may creep up after free trials end. Features may become tied to one office suite, one phone operating system, or one cloud account. Your data may also become harder to move.
For startups, the pressure is harsher. Many AI companies are really wrappers around larger models. That is not an insult. Some wrappers solve real workflow problems. But if the model provider changes pricing, rate limits, or terms, the smaller company can get squeezed overnight (and investors know it).
Independent researchers face a different wall. Closed models can be tested from the outside, but only up to a point. Without audit access, it is hard to verify training data claims, bias controls, security risks, or copyright exposure. Public trust then rests on corporate reports and handpicked benchmarks.
Why Open Source Helps, But Does Not Fix the AI Cartel Behavior Problem
Open-source models from Meta, Mistral, Stability AI, and research groups have changed the debate. They give developers more room to experiment and reduce dependence on a single API. In some sectors, smaller open models are already good enough, cheaper to run, and easier to customize.
But open weights do not erase the compute bottleneck. Training and serving strong models still require hardware, energy, engineering skill, and security work. A small hospital or local newsroom cannot compete with hyperscale infrastructure just because a model is downloadable.
There is also a governance gap. Open systems can improve transparency, yet they can also spread capabilities that bad actors use. Closed systems can reduce some abuse, yet they can hide commercial bias and market control. The smart answer is not “open good, closed bad.” It is targeted oversight based on risk, scale, and market power.
What Regulators Should Do Next
Europe’s AI Act gives regulators a starting point, especially for high-risk systems and general-purpose AI. The United States has taken a more fragmented path, with executive actions, agency guidance, state bills, and antitrust interest from the Federal Trade Commission and Department of Justice. That patchwork leaves gaps big enough for dominant firms to walk through.
Governments should focus less on grand declarations and more on market plumbing. Who controls compute? Who gets audit access? Who can test models before public deployment? Who owns the data supply? Those questions decide whether competition survives.
- Mandate independent audits for the largest general-purpose AI systems, with protected access for qualified researchers.
- Review exclusive partnerships between model labs, cloud providers, chip firms, and major distribution platforms.
- Require data portability so businesses can move prompts, embeddings, logs, and fine-tuning assets where feasible.
- Use public procurement wisely by avoiding single-vendor AI lock-in for schools, courts, hospitals, and agencies.
- Fund public-interest compute for universities, nonprofits, and smaller firms that cannot rent their way into the race.
Procurement may sound dull, but it matters. If public agencies hand long contracts to one or two AI vendors, they help build the very concentration they later complain about. Better contracts can require interoperability, audit trails, and exit plans from day one.
The Real Test Is Who Gets a Seat at the Table
Benanti’s warning should not be read as an anti-tech sermon. It is a demand for accountability before the market calcifies. Once AI systems become embedded in search, office software, hiring, education, health care, and government services, undoing bad power structures will be painful.
The next phase of AI policy should treat concentration as a safety issue. Not the only safety issue, but a non-negotiable one. A market run by a few labs can normalize weak transparency, narrow product choices, and private rulemaking at public scale.
So the practical next step is simple: ask every major AI deal who gains control, who can inspect the system, and how users can leave. If those answers are vague, the cartel warning is already doing its job.
