AI pricing and algorithmic collusion under the Commerce Act 1986
The most detailed guidance is on the topic of pricing algorithms, which has also attracted the greatest attention from competition regulators overseas.
Four ways cartel concerns can arise
It has identified four main ways that an algorithm or AI system can give rise to cartel conduct concerns:
Scenario | How the concern arises |
|---|
Implementing a cartel | Where competitors explicitly agree though humans to coordinate prices, customers, territories, output, or bidding behaviour and then use algorithms to implement, monitor, or enforce that agreement (like the Amazon posters case). |
Shared algorithms and platforms (hub-and-spoke collusion) | Relying on the same independent algorithm, pricing platform, data intermediary, or software provider, which facilitates the exchange of commercially sensitive information between competitors, including through pricing recommendations based on that information (like the RealPage litigation in the United States). |
Sharing sensitive information | Where a business’ own pricing algorithm is trained, calibrated, or operated using current or future competitively sensitive information obtained from competitors. |
Autonomous pricing systems | Where separate systems that are deployed by competitors "self-learn" to interact and produce coordinated market outcomes with each other, which resembles collusion and may occur without knowledge by the business. |
The Commission reminds businesses that section 30 applies also to the fixing of discounts, rebates, allowances, credits and other components of price by an algorithm or AI system.[1]
Our view: unresolved questions about algorithmic collusion
We have four reflections on this guidance:
Sensitive information and the need for an agreement
Like overseas regulators, the Commission has focused on the exchange of competitively sensitive information as the primary vector for collusion. Its guidance, however, does not say that the exchange of such information is necessary to establish cartel conduct, and hints at the potential for broader liability from "algorithms that reduce uncertainty about how competitors will behave". Even then, the Commission would still need to prove the existence of a contract, arrangement or understanding.
When information sharing may complete the conduct
Although the Commission does not specify, it would seem the third kind of cartel (facilitated through sharing of sensitive information) may, depending on the circumstances, be complete once the competitively sensitive information is shared rather than requiring use of the algorithm.
Where historic data becomes current
The Commission only addresses use of "current or future" pricing, output or other data, appearing to accept that use of historic data is permissible. The line between historic and current will vary by market. For example, the Department of Justice's settlement with RealPage, which relates to the residential rental market, prohibited the use of data up to 12 months old.
Autonomous pricing: unresolved questions
The guidance on autonomous pricing systems is the most speculative and uncertain for several reasons:
Limits of the research
First, the Commission refers to research suggesting that advanced autonomous pricing systems "can, under certain conditions", converge towards supra-competitive pricing outcomes. This is heavily qualified for good reason. Those studies are still largely theoretical, and some suggest agent collusion may be fragile in real market conditions.
Establishing an agreement
Second, this theory harm may be conceptually difficult for the Commission to establish because "tacit collusion" or "conscious parallelism" (ie when competing businesses independently mirror each other's market behaviour without any agreement) is not illegal. It is unclear whether the interaction of autonomous pricing systems should always be treated the same as human interaction.
Attributing AI conduct to a business
Third, we also doubt whether the attribution provisions in section 90 of the Commerce Act apply as the Commission seems to suggest, since AI agents (despite their name) are not (at least yet) recognised legal persons. We agree, however, that businesses could, in appropriate circumstances, still be held accountable for the conduct of their AI agents.
Seven guardrails for AI pricing and competition compliance
The Commission has suggested seven "guardrails" for using algorithmic tools and AI systems:
Make decisions independently – AI systems should not be designed or configured to coordinate with competitors, meaning pricing, customer targeting and output decisions must remain independent.
Prohibit use of competitors' sensitive information – This means pricing system should not be trained or calibrated using competitors' confidential pricing, customer, output, or strategic information. Controls should also be in place to prevent inadvertent use.
Update compliance programmes – Staff should be trained to recognise and report potential issues, including understanding how AI can produce anti-competitive outcomes and how to spot unexpected algorithm behaviour.
Check third-party tools carefully – Before adopting pricing software, ensure you understand what data the system uses, whether competitor information contributes to outputs, how recommendations are generated, whether data from different businesses is separated, and what safeguards are in place to prevent anti-competitive outcomes.
Maintain human oversight – Staff should understand how systems work, so they can meaningfully review and challenge, rather than simply rubber-stamping, recommendations.
Conduct regular compliance audits – Periodically review algorithm design and objectives, data sources, pricing outcomes, interactions with competitors, information-sharing arrangements, and compliance controls. Pay particular care to autonomous systems and systems authorised to increase prices.
Keep good records – This includes system design and governance, data sources, competition law risk assessments, audit findings, and pricing tool adoption / modification decisions.
The Commission recommends that if you are concerned about potential competition law issues, you should seek legal advice. For more detailed analysis of algorithmic collusion, see our previous insight on AI-driven pricing and New Zealand competition law, which explores how the issue has been treated by US courts and how those learnings translate to the Commerce Act.