Businesses in all sectors are using artificial intelligence (AI) to improve efficiencies and increase competitiveness. Dentons is no exception, having recently launched FleetAI, its own client secure version of ChatGPT that will empower its lawyers to apply generative artificial intelligence on active client matters.

We explore below an aspect of AI that is getting less attention than some others, but which nonetheless all companies should recognise – competition law risk associated with the use of algorithms (which also encompasses AI applications such as those powered by machine-learning (ML)). One of the co-authors, Cansin Karga, based in our Glasgow office, has recently co-moderated a roundtable session on this issue at the IBA Annual Conference in Paris, which is the biggest legal gathering of the year. 

While AI/ML algorithms have obvious commercial benefits, they also pose novel and perhaps significant risks. Competition law may not be top of the list of these risks for companies, but it can be just as important as data protection, product safety and other more recognised issues, depending on the particular functionality and use of the AI. Pricing algorithms are particularly high risk.

In the same way as companies need to ensure compliance with competition law by employees, companies may now need to start thinking about how their AI is used, or face liability for competition law infringements facilitated by AI technology. 

Competition authorities are clear that companies cannot hide behind algorithms or other types of AI. The European Commission's view is stark: "[B]usinesses also need to know that when they decide to use an automated system, they will be held responsible for what it does. So they had better know how that system works."  

In the UK, the CMA is also looking carefully at AI models and their potential effects, and recently published proposed principles to guide competitive AI markets and protect consumers. 

What are the risks?

Collusion

Pricing algorithms – algorithms that use price as an input and/or a computational procedure to determine the price as an output – have been under scrutiny by competition authorities for a while now. Software to track competitors' prices, and in response adjust a company's own prices, is widely used, particularly in e-commerce. Many platforms offer pricing algorithms to their supply-side users (e.g. Amazon Marketplace, Airbnb). 

Although pricing algorithms can generate pro-competitive effects, such as improving businesses' decision-making and allowing them to make faster and better decisions or helping them to develop more customised services, the main concern is that they may be used to facilitate collusion between competitors. 

Key risks which have been identified by competition authorities are:

• competitors using pricing algorithms to monitor and adjust their prices; and
• competitors subscribing to the same third-party pricing tool using commercially sensitive information from competitors (e.g. future prices) that could result in an unlawful exchange of information.

More controversially, it is theoretically possible that ML algorithms could autonomously learn to collude without human intervention if they come to a common understanding that collusion is the optimum outcome through processing data and communicating with each other. 

The few enforcement cases to date have involved (simple) pricing algorithms used to enforce collusive agreements entered into by individuals, such as in Trod Ltd/GB Eye Ltd, where the parties trading on Amazon Marketplace agreed not to undercut each other and used automated repricing software to monitor and adjust prices. The CMA fined Trod Ltd whilst GB Eye Ltd received immunity as a whistleblower. The Trod director also faced criminal prosecution in the US. In another case, the CMA fined Casio £3.7 million for resale price maintenance implemented by Casio monitoring online prices in real time using different techniques, one of which was the use of price monitoring software called Price2Spy. However, to date there have been no cases involving autonomous collusion based on ML algorithms. 

Market power

Companies with market power might also try to use algorithms to shut out competitors, such as by programming the algorithm to favour their own products and services. There have already been enforcement cases by the Commission for this type of behaviour, with significant fines being imposed. 

Monitoring of the use of algorithms has also been made easier by the EU's Digital Markets Act, which provides the European Commission with powers to request algorithms, data about testing and explanations about their use from companies designated as "gatekeepers" under the Act.

Risk mitigation

It is imperative that companies know how their algorithms work. This should include as a minimum:

• collaboration between legal and technical teams to conduct a risk assessment, including to understand the capabilities of the AI and the sources of information it relies upon;
• creation of a compliance plan (that can be part of a wider competition compliance policy);
• ensuring awareness of the compliance plan by the relevant teams (e.g. procurement); and
• finding alternative solutions for high-risk areas (e.g. switching to an alternative provider to avoid hub-and-spoke collusion if the solution is also used by competitors).

The use of algorithms will continue to grow and competition law enforcers will become more sophisticated in tackling the risks posed. Businesses need to be proactive now and put in place the procedures to identify and deal with evolving risks and issues. Taking the compliance of algorithms with competition law seriously will help avoid burdensome competition investigations in the future or, far worse, fines up to 10% of aggregate group worldwide turnover, costly damages actions and director disqualification.