Applied Case

Applied Case: For the Spicy Chicken

McDonald’s and the AI Exposure Effect

McDonald’s has removed the spicy chicken nuggets from the menu near me.

Again.

The McRib continues to appear and disappear according to some mysterious corporate calendar somewhere. Apparently these meats have entered some kind of sealed interpretative system. I would like them returned to us.

This was initially going to be a complaint about some mediocre chicken.

Then McDonald’s got sued over artificial intelligence.

And, unfortunately for McDonald’s, the spicy chicken turned up in the evidence.

On February 12, 2021, Reuters reported that McDonald’s franchisees were using Deloitte to advise them on menu pricing. One franchisee described Deloitte’s recommended prices for the chain’s new chicken sandwiches:

$3.69 for the spicy and crispy versions.

The franchisee intended to charge $3.99 instead, citing higher local wages.

There it is.

Five years before the current lawsuit, a professional pricing adviser was calculating what the spicy chicken should cost, and a franchisee was deciding whether to follow that advice.

This is an excellent place to begin.

Because the question now being asked about McDonald’s artificial intelligence has a history much older than the machine-learning system currently attracting attention.

Who actually controls the price?

And, more importantly, when does advice become control?

The Lawsuit Arrives.

On October 2, 2026, Michael Thomas filed a proposed nationwide class action against McDonald’s USA and McDonald’s Corporation in the United States District Court for the Northern District of Illinois.

Thomas v. McDonald’s USA, LLC, No. 1:26-cv-12149.

The 31-page complaint alleges that McDonald’s operates a centralized machine-learning pricing engine using nonpublic information from independently operated restaurants. According to Thomas, the system facilitates unlawful coordination among franchisees that should be competing with one another.

The complaint invokes the Sherman Act, Illinois antitrust law, and Illinois consumer-protection law. It seeks damages and injunctive relief.

These are allegations. The proposed class has not been certified, and the lawsuit has not established liability.

McDonald’s denies the underlying accusation.

On October 1, the company published a statement titled Separating Fact from Fiction: AI Does Not Set Prices at McDonald’s.

Its position is pretty clear.

Franchisees set their own menu prices. The company's recommendation tools provide information.

McDonald’s says the tools cannot automatically set or change prices, that individual customers are not charged personalized prices, and that the company does not use real-time dynamic pricing.

Those distinctions do matter.

A restaurant-specific recommendation is different from charging two customers different prices because a model estimates their individual willingness to pay. And neither arrangement automatically establishes an antitrust violation.

The question is whether the independently operated restaurants actually make independent pricing decisions, including when corporate headquarters disagrees with them.

To investigate that, we need to go backward.

Quite a bit further than ChatGPT.

So back in January 2007, economists Itai Ater and Oren Rigbi published a Stanford working paper:

Price Control in Franchised Chains: The Case of McDonald’s Dollar Menu.

They examined restaurant-level price data from 1999 and 2006, spanning McDonald’s introduction of the Dollar Menu in 2002.

They found that the price premium charged by franchise-owned restaurants relative to company-owned restaurants had fallen by approximately 70%.

Importantly, that reduction was concentrated in products with close substitutes on the Dollar Menu.

Think about what that means.

McDonald’s did not have to directly specify the price of every hamburger to influence what its franchisees could profitably charge.

A nationally promoted cheap item changes the commercial environment surrounding other items. A local restaurant may technically retain the authority to set a higher price. Customers retain the ability to buy the cheaper substitute.

The pricing choice has changed before any order arrives telling the franchisee what number to enter.

That is corporate influence through menu architecture.

The study does not establish an unlawful agreement, and it does not show that McDonald’s was operating a machine-learning pricing engine under our noses in 2002.

It establishes something more useful for our present question.

The corporation already possessed instruments capable of influencing franchise pricing without directly setting every franchise price.

The machines came later.

The incentive to govern prices through the surrounding system was already there.

The Franchisees Warned Everyone.

On October 15, 2018, McDonald’s franchisee Blake Casper published a warning as hundreds of operators moved to establish the National Owners Association.

He argued that corporate control had been migrating from local owners toward national headquarters.

And he named the next instrument.

The next control change will be the transfer of Menu Board Pricing to Deloitte and Touche.

That stark warning was reported by Meat+Poultry on October 17, 2018.

Casper argued that experienced restaurant owners were better positioned to balance local costs, customers, and market conditions than distant financial advisers.

There is a perfectly respectable debate to be had here.

Local operators can misprice products. Corporate analytics can improve pricing. National promotions can bring customers back. Franchisees can also raise prices too aggressively, weakening the brand and reducing customer traffic.

The dispute becomes more consequential when one side controls the other side’s ability to continue operating.

Casper's warning did not prove that a transfer to Deloitte had already occurred. It recorded a franchisee's public concern about the future distribution of pricing authority.

That concern was on the record in 2018.

Artificial intelligence did not have to invent it.

The Models Were Already Here.

It is important to get the technology right.

Machine learning did not begin with large language models. Artificial intelligence did not suddenly become available to businesses in November 2022.

McDonald’s was investing in predictive systems long before that.

In March 2019, McDonald’s announced its acquisition of Dynamic Yield, describing plans for personalized drive-through displays and product recommendations based on factors such as time of day, weather, traffic, and ordering patterns.

McDonald’s subsequently sold Dynamic Yield to Mastercard, with the acquisition completed in April 2022.

That system concerned ordering personalization. Modal Path Ethics has not established that Dynamic Yield operated the pricing engine challenged in the present lawsuit. Keep the systems separate.

Then, in February 2020, Deloitte Digital published How Strategic Pricing Is Changing the Restaurant Industry.

Deloitte described a Strategic Pricing Analyzer using artificial intelligence and machine learning to combine restaurant sales, costs, competitive data, and other inputs into predictive pricing recommendations.

It described a recommendation portal, pricing advisers, and organizational processes for getting operators to adopt the system.

Deloitte said the technology had been developed with a leading restaurant company. The publication does not identify that company, so we cannot establish that the described installation was McDonald’s.

But the technology and its accompanying management model were already publicly documented.

And by February 2021, Reuters had independently documented Deloitte advising McDonald’s franchisees on menu prices.

Including the spicy chicken.

The evidence does not give us one continuous, verified software lineage from Deloitte's 2020 product description through McDonald’s current Tiger Analytics platform.

It does establish a historical succession of

corporate pricing influence,
professional pricing advice,
machine-learning systems, and
increasingly explicit pricing oversight.

The public controversy may be arriving pretty late to the party.

The AI Exposure Effect.

Modal Path Ethics has encountered versions of this problem before.

In AI Did Not Blur Reality, the problem was a social-media environment that had already been translating real events into engagement-driven representations before generative artificial intelligence could manufacture those representations directly.

In Tales of Distortion: The Arrival of Artificial Intelligence, the question was what happened when new cognitive capacity entered institutions whose objectives, incentives, and dependencies already existed.

McDonald’s offers a new and unusually concrete version.

Call it the AI Exposure Effect.

The AI Exposure Effect occurs when the introduction, expansion, or public scrutiny of an artificial-intelligence system makes an older institutional mechanism newly visible, while public attention risks attributing the mechanism’s origin to the technology that just exposed it.

This effect has two sides.

First, artificial intelligence can make the existing mechanism more powerful.

It can process more data, produce more granular recommendations, coordinate activity across more sites, reduce the time needed for decisions, and create detailed records of adherence or deviation.

These are real technical changes.

Second, the new technology can make the surrounding institutional arrangement visible.

A practice previously described as ordinary management becomes a question about who supplied the data, what the optimization target was, which businesses received recommendations, and what happened to people who refused them.

The model becomes the object everyone wants to investigate.

And the investigation reveals a system that was already operating.

There is a trap here.

If we respond by blaming artificial intelligence for inventing the whole arrangement, we may leave the arrangement intact.

Replace the machine-learning system with a spreadsheet, a consulting firm, or a room full of analysts, and the same underlying distribution of power might survive.

Conversely, recognizing the older mechanism does not excuse the new technology.

A system that was previously limited by human attention can become materially different when computation removes those limits.

The exposure effect is therefore a diagnostic claim about history, attribution, and changed capacity, not a universal law, an assertion that all older practices were harmful, or an exemption from responsibility.

We must establish what existed before, what changed, and what that change allowed the institution to do.

McDonald’s gives us spicy evidence for that investigation.

When the Recommendation Starts Watching You.

On December 8, 2025, McDonald’s announced enhanced franchising standards, effective January 1, 2026.

Its own supplemental explanation says franchisees would continue making their pricing decisions with third-party advisers. Corporate headquarters would assess the outcomes for customer value.

McDonald’s also acknowledged that value had been part of franchise business reviews for years.

That detail prevents us from manufacturing a clean before-and-after story.

January 2026 was an identifiable policy change. It was not necessarily the first moment corporate headquarters could influence pricing through franchise reviews.

Then Reuters published its September 29, 2026 investigation.

The reporting drew on pricing-engine screenshots, company documents, and interviews with nine people familiar with McDonald’s strategy.

It found that the machine-learning engine generates restaurant- and item-specific recommendations from millions of daily transactions.

Five restaurant owners described corporate pressure to use the pricing tools. Reuters also examined a June document recording franchisees’ deviations from recommended prices.

McDonald’s disputes the characterization that the tools are compulsory.

But the important question now becomes sharper.

What happens when a franchisee rejects the recommendation?

One refusal might produce nothing.
Another might prompt a conversation.
Repeated deviation might affect a business review.
And the business review can influence whether a franchisee remains eligible to expand or renew.

On an August 2026 earnings call, CEO Chris Kempczinski acknowledged that pricing execution and noncompliance could enter franchise business reviews affecting growth and eligibility.

The immediate discussion concerned a national value-menu program, including items priced below $3. That is not proof that rejecting a recommendation from the separate machine-learning engine triggers a franchise penalty.

The distinction is essential.

What we have here is documented corporate oversight of pricing outcomes and reported monitoring of algorithmic pricing recommendations. The precise connection between those systems needs further examination.

A recommendation does not acquire binding power just because someone calls it intelligent.

It acquires practical power through the institution receiving it.

If rejecting an ostensibly optional recommendation predictably threatens a franchisee’s ability to continue operating, then the freedom to reject has become materially thinner than its formal description suggests.

The decision remains legally assigned to the franchisee.
The conditions surrounding the decision may have moved elsewhere.

That is the crossing to locate.

30% of the Restaurants.

Reuters reported something particularly interesting about the pricing engine.

According to two former Tiger Analytics employees, McDonald’s supplied rules and corporate targets that influenced its recommendations.

One rule reportedly limited certain recommended price increases to items whose prices had already risen recently at at least 30% of stores.

Another directed attention toward items that had gone two years without a price increase. Other parameters excluded certain summer items from increases.

The first rule deserves an audit.

If the recommendation made to Restaurant A depends upon price changes previously made by competing Restaurants B, C, and D, then the system contains a cross-restaurant pricing dependency.

The current lawsuit treats that dependency as evidence of an alleged coordination arrangement.

But the rule by itself does not establish unlawful price-fixing.

The engine might use such a threshold to avoid premature increases. It might produce legitimate competitive guidance. The decisive questions concern the underlying information, the agreements among participants, the actual recommendations, and the consequences for competition.

The public account does not establish all of those facts.

Still, there is a question worth putting directly to the model and its operators:

Would Restaurant A have received the same recommendation if its competitors had not raised their prices?

Then ask whether Restaurant A could safely refuse.

Those are two separate investigations.

One concerns coordination through information.
The other concerns institutional control through dependency.

Together they identify the actual architecture.

The Computer: Not the First Conspirator.

Antitrust law encountered computerized pricing coordination long before large language models.

In December 1992, the United States Justice Department sued eight airlines and Airline Tariff Publishing Company, alleging that airlines used a shared computerized fare system to communicate future price changes and coordinate their behavior.

The resulting case produced consent decrees in 1994.

Then, in 2015, the Justice Department announced a criminal prosecution involving online poster sellers who had agreed to coordinate prices using algorithms.

No chatbot was necessary.

The legal problem was the agreement and the competitive conduct, whatever machinery carried them.

In March 2024, the Justice Department and Federal Trade Commission made the same point in an algorithmic hotel-pricing case. Their statement of interest explained that competitors cannot necessarily escape antitrust scrutiny by coordinating recommended starting prices while retaining some discretion over final prices.

That remains an agency legal position, not a ruling that every shared pricing tool violates the Sherman Act.

McDonald’s has an important defense available.

Franchises belong to a common brand. National promotions, pricing advice, brand reputation, and customer expectations create legitimate reasons to coordinate some activities.

And courts have recognized that franchise relationships can justify restraints serving the shared business.

What remains disputed is whether this particular arrangement crosses from legitimate franchise coordination into unlawful competition restraint.

The word algorithm does not settle that question.

Neither does the word independent.

The Machine Sometimes Wants Cheaper Burgers.

There is a fact that makes the simplistic version of this story collapse.

Reuters reported that McDonald’s pricing engine has recently recommended price decreases as well as increases.
Some franchisees resisted those lower prices because their own restaurant costs were rising.

Corporate headquarters and franchise operators do not always have identical incentives.

McDonald’s earns revenue partly through royalties connected to franchise sales. A strategy that attracts more customers at lower prices can improve system-wide sales while placing pressure on individual franchisees' margins.

A franchisee might want to charge more to cover wages and operating costs.
A customer might benefit from the lower price.
The corporation might benefit from higher sales volume.
A worker might be affected by the operator’s response to narrower margins.

The machine can optimize accurately within that arrangement and still tell us very little about the arrangement’s overall moral quality.

Way back in Field Instruments: Money, Modal Path Ethics argued that prices are useful signals from inside a field. They do not constitute the field’s moral verdict.

A price recommendation can be financially accurate. It can also be competitively lawful. Those are important achievements. Neither guarantees that all affected continuations have been preserved or that every burden has been counted.

And this cuts both ways.

McDonald’s own May 2024 pricing fact sheet attributed its average menu-price increases of roughly 40% since 2019 to substantially increased restaurant input costs, including wages and food. The observed increase does not prove the pricing engine caused an unlawful overcharge.

But aggregate cost inflation cannot independently establish that every algorithmic pricing practice was competitive, either.

We need comparison against the actual alternative pricing paths.

Which prices would independent restaurants have selected with their own data and legitimate outside information, without any unlawful coordination?

What the Exposure Actually Revealed.

We can now identify the historical movements without pretending that the record is complete.

In 2002, corporate menu design already supplied a mechanism for influencing franchise prices.
By 2018, franchisees were openly worried that further pricing authority would shift toward national management and outside advisers.
In 2020, machine-learning restaurant pricing systems with recommendations and adoption processes were publicly documented.
In 2021, Deloitte was advising McDonald’s franchisees on the price of spicy chicken sandwiches. At least one franchisee intended to reject the recommended price.
In 2026, Reuters documented a much more elaborate machine-learning pricing infrastructure, deviation tracking, and pressure reported by franchisees. McDonald’s also had enhanced pricing oversight through its franchising standards.
- Then the lawsuit arrived.

The current record supports the existence of an older pricing-control problem, a later computational expansion, and a new public controversy focused on artificial intelligence.

It does not yet establish the precise date when advisory pricing, if ever, crossed into unlawful coordination. Nor can we infer that the introduction of machine learning itself caused that crossing.

The real threshold may lie in the data exchange. It may lie in the agreement to use recommendations. It may lie in a later enforcement practice. There may be several distinct transitions, only some of which create legal liability.

Or the disputed conduct may ultimately be found lawful.

Those possibilities remain available. That is what makes this an investigation rather than a verdict manufactured out of a headline.

The AI Exposure Effect gives us a better method for conducting it.

When a new artificial-intelligence scandal appears, recover the older institution.
Find the preexisting objective, the data arrangement, the decision rule, the material dependency, and the people who already had reason to object.
Then establish what exactly the new technology changed.
Finally, identify which of those changes matter to the actual harm or legal question.

Artificial intelligence can make a corporation more effective at doing something good, harmful, lawful, unlawful, or mixed.

The age of the objective does not excuse the latest implementation.

The sophistication of the implementation does not establish that the objective began with the machine.

And the sudden visibility of the machinery does not mean the machinery suddenly came into existence.

That is the AI Exposure Effect.

Ruling: For the Spicy Chicken.

Modal Path Ethics cannot currently conclude that McDonald’s illegally fixed prices.

The lawsuit is pending. The key records are disputed or unavailable for independent examination. The historical evidence supports a serious inquiry into control over franchise pricing, and Reuters has made that inquiry substantially more concrete.

We can reach a separate conclusion about the technology.

The AI headline arrived after the institution.

McDonald’s already possessed mechanisms for influencing local prices, franchisees had already organized against growing central control, pricing advisers were already producing recommendations, and machine learning had already entered restaurant economics before the current generative-AI era.

The new system may have intensified those relations. It certainly provides new evidence through which to investigate them.

That is the Exposure.

The point of identifying it is to keep the investigation aimed at the whole causal arrangement.

Replacing one model will not necessarily change the incentives that commissioned it, the authority that controls the franchise, or the rules by which an independently owned restaurant is judged.

Correcting the arrangement, where correction is warranted, requires reaching those relations.

Now.

There is one final matter.

McDonald’s has assembled a machine-learning system capable of processing millions of transactions and generating restaurant-specific pricing recommendations.

In February 2021, its pricing advisers were already calculating exactly what to charge for spicy chicken.

Five years later, I am sitting here in the parking lot without any spicy chicken.

I recognize that seasonal menu availability and algorithmic price coordination are separate commercial questions. I am making no allegation that the absence of the sandwich constitutes an antitrust violation.

I am also making no promises about the limits of this investigation.

Bring back the spicy chicken.

Keep it all year.

And the McRib.

You have had decades to work out the economics here.

Modal Path Ethics is prepared to pay a reasonable price.

Sources

Case status: allegations disputed; no finding of McDonald’s antitrust liability established as of October 7, 2026.

MULTIPLAYER: Abdul El-Sayed and the 50/50

Russell T Davies and the Writer Who Was Always There

IGN Has Mistaken a Cool Crocodile for an Argument