AI Regulation Tracker / Antitrust litigation
A shared AI pricing algorithm can now supply the rim of a price-fixing conspiracy
Any client that feeds current, non-public pricing or occupancy data into a vendor AI engine its competitors also use needs its revenue-management policy re-read this week. A federal court of appeals has just held that pleading that is enough to survive dismissal.
What did the Third Circuit actually decide, and what did it not decide?
The district court dismissed the consolidated amended complaint because the plaintiffs had not pleaded the rim of a hub-and-spoke conspiracy, an agreement running between the casino-hotels themselves. The Third Circuit disagreed. Its operative sentence is short: "We hold that the well-pleaded allegations in the CAC are sufficient to support a finding that casino-hotel Defendants have conspired to fix prices through Cendyn's software."
Read that sentence carefully. Sufficient to support a finding is not a finding. The panel was reviewing a Rule 12(b)(6) dismissal, so it took the allegations as true. It said the plausibility standard "is not a probability requirement," and added a footnote that at the next stage plaintiffs "will face a higher burden to sustain their claims by further developing the facts of the case." The case goes back to the District of New Jersey.
The court was also explicit that it was not condemning the technology. It wrote that "there is nothing inherently wrong [or anticompetitive] with using [algorithms] to engage more effectively in commercial activity," and said it made no assumptions or conclusions about how the software actually works.
Which facts did the court hold sufficient to plead the rim?
This is the part that matters operationally. The panel did not accept a general theory that shared software equals conspiracy. It accepted a specific stack.
- Each casino-hotel continuously supplied its current, non-public room pricing and occupancy data to the vendor platform, allegedly installed directly into on-site pricing systems.
- The algorithm processed that data along with the same type of non-public, real-time data submitted by participating competitors, then generated suggested rates for each client, updated multiple times a day.
- Each defendant allegedly knew the rates it received were built from non-public data that it and its co-defendants were all providing, and knew its competitors understood the same thing. Mutual awareness, pleaded on both sides.
- Defendants allegedly charged the recommended rate about 90 percent of the time, with overrides requiring special permission for times of need and extreme circumstances, and with each client scored on how often it overrode.
- Parallel conduct plus plus-factors: motive from years of financial hardship in a concentrated market, conduct against self-interest, and non-economic evidence of a traditional conspiracy including industry events and a sudden break from historically independent pricing.
On the economics, the complaint alleged that between 2017 and 2019 collective occupancy fell about 8 percent while room revenue rose about 22 percent, and that in 2022 the casino-hotels rented 5 percent fewer rooms while charging 25 percent more than in 2019. The court treated a refusal to cut rates while occupancy slid as against self-interest here, where rooms are the funnel and the casino floor is the revenue.
Two defenses did not carry the day at this stage. First, staggered adoption. The defendants adopted the platform at various points across a fourteen-year window, and the court said the alleged conspiracy did not begin with adoption but with continuous parallel conduct during the class period. Second, retained pricing authority. It was undisputed that the hotels could override the suggested price. The court answered with Masonite: "Prices are fixed when they are agreed upon," regardless of whether conspirators always adhere to them.
If it isn't ok for a guy named Bob to do it, then it probably isn't ok for an algorithm to do it either.Former FTC acting chair Maureen K. Ohlhausen, quoted in the opinion, No. 24-3006 at 42
Who owes a duty on Monday, and what is the duty?
The duty falls on counsel and compliance leadership at any firm that contributes competitively sensitive, non-public data to a pricing, yield or revenue-management tool that the same vendor runs for the firm's competitors. That is a wider group than hotels: multifamily landlords, healthcare and insurance pricing teams, transport and freight, any e-commerce operation on a shared repricing service. It reaches the vendors too. Cendyn Group is a named defendant here, and the opinion says that whether the software operates as a hub, a facilitating practice, an intermediary or a conduit, if it is plausibly alleged to facilitate a horizontal agreement among some of its clientele it crosses the line under the court's Sherman Act jurisprudence.
The practical duty is to know, and to evidence, what data leaves the building, whether it is non-public and current, and whether the output coming back is derived from competitor inputs. Most revenue-management contracts do not answer those questions on their face. Ask the vendor in writing.
The second duty is documentary. The opinion notes that there are rarely legitimate business justifications for affording competitors the benefit of commercially sensitive proprietary information under the circumstances alleged there. An independent business reason for a data feed should exist in a contemporaneous record, not in a declaration drafted after a complaint lands.
How does this sit against the other algorithmic-pricing entries on this tracker?
We track three other lines of attack on algorithmic pricing, and they turn on different triggers. Our entry on the DOJ consent decree with Willow Bridge over RealPage rent pricing covers the enforcement route, on the same axis as this opinion: competitors' non-public data inside a vendor algorithm. Our entry on the New Jersey FAIR Act banning algorithmic rent-setting covers the legislative route in one sector. Our entry on New Jersey's Fair Price Protection Act on surveillance pricing is the odd one out: it targets personal-data-driven individual pricing of consumers, not coordination among sellers. Confusing the two is the most common error here.
| Route | What triggers exposure | Non-public competitor data required | Where it stands |
|---|---|---|---|
| Cornish-Adebiyi, 3d Cir. No. 24-3006 | Pooled non-public inputs, returned recommendations, mutual awareness, high adherence | Yes, central to the holding | Dismissal reversed 29 July 2026, remanded. Pleading stage only. |
| Gibson v. Cendyn Group (as cited in the opinion) | Same vendor. That district court found the theory did not cross from conceivable to plausible | Yes, and found insufficiently pleaded | D. Nev. dismissal 8 May 2024, affirmed 148 F.4th 1069 (9th Cir. 2025), certiorari denied 20 April 2026 |
| DOJ consent decree, Willow Bridge (our earlier entry) | Use of competitors' non-public data in a rent-pricing algorithm | Yes, the express object of the bar | Proposed decree filed 6 July 2026 |
| New Jersey FAIR Act (our earlier entry) | Statutory ban on algorithmic rent-setting in housing | No, the ban is drafted by conduct, not by data type | Passed 30 June 2026, sent to the Governor, recorded as not yet enacted |
One caution on the second row. The Third Circuit cited the Nevada case only as the authority the district court had leaned on. We are not characterising the Ninth Circuit's reasoning, because we did not open that opinion, and we are not calling this a circuit split.
What did the court refuse to decide?
Three things. Whether the alleged restraint is per se unlawful or subject to the rule of reason on a developed record. Whether the district court abused its discretion in denying leave to amend, which reversal made moot. And how the software functions, the court saying that requiring plaintiffs to plead the workings of proprietary software before discovery was neither required nor appropriate at this stage.
The court did take seriously the appellees' amicus warning against criminalising industry-wide use of the same software. It answered the shared-spreadsheet hypotheticals by saying they oversimplify, and pointed back to the pleaded facts. That is the boundary line. Common software, standing alone, is not the theory. Common software fed with competitors' current confidential numbers, returning prices the participants follow, is.
What belongs in a shared-algorithm policy now?
A working checklist, not legal advice on your facts. Inventory every pricing, yield and revenue-management tool and identify whether the vendor also serves direct competitors. Classify the outbound data as public or non-public, current or historical, because the opinion turns on current and non-public. Get contractual clarity on whether your inputs are pooled with competitor inputs. Record override rates and reasons, because adherence percentages are now a pleaded metric. Ask whether the vendor scores clients by adherence. And check what your executives say at industry conferences about avoiding price wars, because the complaint quoted a vendor executive on avoiding a race to the bottom and the court found it significant.
Frequently asked questions
Did the Third Circuit find that Caesars or Cendyn fixed prices?
No. This is a pleading-stage ruling on a Rule 12(b)(6) motion. The court took the well-pleaded allegations as true, held they were sufficient to support a finding of conspiracy, reversed the dismissal and remanded. No defendant has been found liable, and a footnote notes that at the next stage plaintiffs will face a higher burden by further developing the facts of the case.
Did the court hold that using a common pricing algorithm is illegal?
No. The opinion states that there is nothing inherently wrong or anticompetitive with using algorithms to engage more effectively in commercial activity, and it rejects the amicus hypotheticals about shared spreadsheets and forecasting software as oversimplifying. What mattered was the combination alleged: competitors continuously feeding current, non-public pricing and occupancy data into one vendor platform, recommendations built from that pooled data and returned to each, alleged mutual awareness, and alleged acceptance of the recommended rate about 90 percent of the time.
Does the ruling bind companies outside the Third Circuit?
It is binding precedent only in the Third Circuit, which covers Delaware, New Jersey, Pennsylvania and the Virgin Islands. Elsewhere it is persuasive authority. The opinion records that the district court had relied on Gibson v. Cendyn Group, No. 2:23-CV-00140 (D. Nev. May 8, 2024), affirmed at 148 F.4th 1069 (9th Cir. 2025), certiorari denied April 20, 2026.
Does keeping final pricing authority protect a company?
Not by itself, at the pleading stage. It was undisputed that the casino-hotels retained final pricing authority and could change the suggested price. The court held that this did not defeat the claim, quoting Masonite that prices are fixed when they are agreed upon, and pointing to the alleged 90 percent acceptance rate and override permissions limited to select staff.
A note on what we have not said. This is not a decision that algorithmic pricing is unlawful, and the casinos have not lost. We have also declined to name the former vendor executive quoted in the complaint, because the court identifies that person only by role.
Last verified: July 29, 2026