Back in September 2023, which in terms of innovation would be considered an eternity ago, the Competition and Markets Authority (“CMA”) published its initial review of AI Foundation Models (“FM”). FMs are large-scale machine learning models pre-trained on extensive data sets which enable them to be fine-tuned for a wide variety of specific tasks. The initial report set out potential risks to competition and consumer protection and introduced a proposed set of guiding principles. You can find out more about the initial report in our commentary here. Fast forwarding to April 2024, the CMA published an update paper where it details its “concerns that the FM sector is developing in ways that risk negative market outcomes”. 

Whilst the CMA does not anticipate FMs bringing about the end of humanity, it has become increasingly concerned about the fairness, openness and effective competition in this space, particularly given the growing presence of small number of technology firms that already hold positions of power in key digital markets. The update paper sets out the actions CMA is currently taking and will consider taking to address the risks.

Key developments since the initial report

The use of FMs continues to rise, with 31% of adults and 79% of 13-17-year-olds in the UK using generative AI (most, but not all generative AI models are FMs) in either a personal, educational or professional capacity. According to the Office for National Statistics, 15% of UK businesses use generative AI, and this number increases to 46% for businesses with more than 250 employees. 

Just as the number of users is rising, so is the number of FMs and their capabilities. The CMA has identified a nearly threefold increase in the number of FMs, now totalling 330. Those FMs are now able to produce text, audio, images and video outputs and process increasing volumes of data. 

While the availability of accelerator chips remains a concern to sustained development, new market players have either released or announced future releases. Computing resources are generally accessed via partnerships with cloud service providers, and there is a growing trend of using synthetic (artificially generated) data, particularly for pre-training and fine-tuning purposes. 

Lastly, and as a result of new information to hand, the CMA updated its AI principles which are listed in the below table: 

Three-headed monster

Technology firms are increasingly active across multiple levels of the FM value chain, from supply of data or compute, to the provision of end products or services. This has been made possible due to an increase in vertical integration and strategic partnerships within the sector. As such, the CMA has identified three key risks, detailed below, among its concerns that these firms can shape the market development to the detriment of competition, choice and innovation.

Firms that control critical inputs for developing FMs may restrict access to them to shield themselves from competition 

CMA is currently examining the competition in provision of cloud services, Microsoft’s partnership with OpenAI, and the competitive landscape in the AI accelerator chips market. As for future plans, the CMA will monitor market developments to select priority activities to investigate under the new powers in the Digital Markets, Competition and Consumers (DMCC) Bill, which may include computing. 

Powerful incumbents could exploit their position in consumer or business-facing markets to distort choice in FM services and restrict competition in FM deployment 

This is demonstrated by examples such as Microsoft deploying its (and its partner OpenAI’s) FMs within Office, or Google using its own FM for its Search Generative Experience. To address this risk, CMA will also consider which digital activities to investigate under their new DMCC powers, listing critical points of access or routes to market such as mobile ecosystems, search or productivity software as potential examples. 

Partnerships involving key players could reinforce or extend existing positions of market power through the value chain 

The CMA has identified an interconnected web of 90 partnerships in the FM value chain between key technology firms such as Google, Amazon, Microsoft, Meta, Apple and Nvidia, together known as GAMMAN, and is uneasy about the possibility of these partnerships supressing competitive threats. 

To address these risks, CMA is:

• Monitoring current and emerging partnerships;
• Stepping up use of merger control to identify more clearly and coherently which types of partnerships fall within the merger rules; and 
• Examining Microsoft’s relationship with OpenAI.

In addition, CMA will also monitor firms’ approaches to monetising FMs, their integration and development into new products and services. Similar to the examples mentioned above, further monitoring will help determine areas of focus within new DMCC powers. 

What’s next?

With high entry barriers to the FM markets and existing CMA concerns, it may be a matter of time before the regulator takes action against GAMMAN domination. Meanwhile, the CMA urges firms to align their business practices with their confirmed AI principles to shape positive market outcomes. 

The CMA acknowledges that there are still many unknowns, not limited to its own powers under DMCC. Therefore, the forthcoming update this Autumn, as along with a paper on AI accelerator chips and a joint statement with the Information Commissioner’s Office will, hopefully, provide us with further guidelines. 

Meanwhile, given how much change FMs have undergone since the initial report, the FM market is likely to continue evolving at pace. The CMA will likely have again, a lot of catching-up to do, by the Autumn update. However, developers and users of FM models should nevertheless remain vigilant as they can expect increased scrutiny from the CMA for the foreseeable future.