Project Nessie

The FTC says Amazon ran a tool that spotted products where rivals would follow a price hike, raised the price, and held it once they matched. Amazon disputes this.

A billion in excess profit

The FTC alleges the tool produced over $1 billion in extra profit, and that Amazon switched it off during scrutiny then back on. Amazon says it was retired years ago.

German gas stations

In 2017, automated pricing software spread to German fuel stations. Where two rival stations both adopted it, margins rose about 38%, with no meeting and no message.

“no meeting, no message, and no agreement between them”

One adopter changes nothing

Market margins stayed flat when only one station used the software. The jump appeared only once two algorithms were left to set prices against each other.

A cartel with no cartel

Pricing software can deliver what a cartel delivers, higher prices held over time, without any of the conduct antitrust law was written to catch.

Calm markets lie now

Executives read steady prices and comfortable margins as a win. With autonomous agents, that same picture can mean competition quietly stopped.

The dangerous success

An obviously wrong price is easy to catch. The hard case is a system doing exactly its job, optimizing margin, and landing somewhere you cannot defend in public.

The ghost

Independent algorithms, each chasing its own profit, learn over repeated rounds to stop undercutting. No agreement, no data shared, no designer.

“just two systems that arrived at the same truce independently”

The mirror and the hub

The mirror is one firm predicting a rival's reaction and moving first, like Nessie. The hub is competitors feeding data to a shared provider, as with RealPage.

Bots learned to punish

In a 2020 American Economic Review paper, reinforcement-learning agents that could not communicate still priced above competitive levels and punished any one that cut.

“delegating pricing to algorithms opens a backdoor to collusion”

The law needs a handshake

Enforcement was built around evidence of a meeting or understanding. Coordination a machine works out alone leaves none of that behind.

RealPage got off light

DOJ sued in 2024 over rent software fed by competitor data. The November 2025 proposed settlement brought no penalty, no admission, just data limits and a monitor.

Where courts drew the line

The Ninth Circuit tossed a case against Las Vegas hotels because no confidential data was pooled. The Third Circuit revived a casino case where it was, and was followed nine times in ten.

Cities banned it first

San Francisco started in 2024, then Philadelphia, Minneapolis and Seattle. New York passed a statewide ban in October 2025, and RealPage sued, calling its recommendations protected speech.

The risk is yours

Even where the legal categories miss, reputational and regulatory exposure lands on the company that deployed the system and set its objective. You cannot outsource that to the vendor.

“that responsibility cannot be outsourced to the vendor”

Audit behavior, not results

Map which systems can see rival prices, replay markets with competitor signals delayed or removed, and document the mandate. If margins collapse without rival data, investigate.

“When prices emerge from rules rather than from individual decisions, those rules are the decision.”