Brazil Targets AI Pricing Tools in Cartel Probes

Brazil's competition regulator is building a comprehensive enforcement framework around AI-powered pricing tools, targeting both the companies that deploy them and the third-party vendors that supply them. Two parallel cases opened in April 2025 by the Administrative Council for Economic Defense (CADE) reveal how antitrust liability is expanding beyond traditional cartel conduct into algorithmic coordination.

Airlines under investigation for algorithm-enabled parallelism

CADE opened an investigation into GOL and Latam, Brazil's major domestic carriers, examining whether the airlines used AI and machine learning tools to indirectly share competitively sensitive pricing information. According to the enforcement agency's technical note, a confidential third-party vendor allegedly served both airlines simultaneously, collecting and processing real-time fare data with contractual authorization to redistribute that information to other market participants.

The investigation theory centers on "conscious parallelism" facilitated by technology rather than direct communication. Notably, CADE is exploring a negligence-based liability theory—culpa in vigilando—that would hold companies responsible for failing to supervise the collusive potential of algorithms they contract, even when the vendor operates the system.

Fuel pricing software vendor settles with commitments

In a parallel case, CADE reached a cease-and-desist settlement with Intelprice, developer of the Aprix pricing software used by gas stations. The investigation found that Aprix collected competitor pricing data, historical sales information, and operational details from client stations, then allegedly used marketing materials instructing clients to resist price cuts and warning about risks from independent pricing decisions.

The settlement requires Intelprice to pay a fine and implement contractual confidentiality clauses with clients, establish an antitrust compliance program, grant CADE facility access for external audits funded by the company, and notify the regulator when serving companies representing 20% or more of any local market. This notification threshold, a CADE official noted at a recent event, aims to assess the vendor's influence over pricing dynamics in specific geographic areas and may vary by market.

Why it matters

These cases establish that antitrust exposure from pricing algorithms doesn't require participation in a traditional cartel. Companies face liability based on what data their contracted tools access, whether those tools contain competitors' confidential information, and how much oversight they exercise over vendor practices. The enforcement pattern also extends scrutiny to how pricing solutions are marketed—a factor that increases exposure for both users and providers. For technology vendors serving multiple competitors, the Brazilian approach creates compliance obligations that resemble those in anti-corruption frameworks, including audit requirements and proactive disclosure duties.

Architectural features that raise red flags

CADE's enforcement priorities are taking shape around specific risk factors. Tools that automatically determine and publish AI-set prices represent the highest-risk configuration. By contrast, systems that merely suggest prices while preserving human decision-making authority appear to carry lower risk, provided companies maintain genuine independence in final pricing decisions.

The presence of competitors' confidential data in algorithm databases emerges as a central concern across both cases. CADE officials indicated that certain forms of algorithmic coordination might be treated as violations "by object"—conduct inherently anticompetitive without requiring proof of actual market harm.

The evidentiary challenges posed by opaque algorithmic systems are prompting CADE to consider shifting the burden of proof to companies in future cases. Currently, economic evidence is being used to demonstrate that problematic conduct occurred rather than to measure market effects, reflecting the practical difficulty of determining what algorithms are doing with competitor data.

These developments were first reported by Wolters Kluwer's Competition Blog, based on case documents and comments from the CADE official handling the Aprix matter.