Foundational Regulation: UK Competition & Markets Authority Outlines Vision for Competition Regulation of AI Foundation Models

Foundation models are very large AI models that differ from traditional AI models not only in their size — often being trained on terabytes of data — but also in their broader usefulness, potentially able to perform a wide variety of tasks including text or image generation, classification, conversation, generating of computer code and more. The CMA Report recounts the history of FM evolution and the technical aspects of their development. The CMA Report describes “competition and barriers to entry” in the development of FMs including data, computational resources, technical expertise, access to funding, and open-source models. The CMA Report also identifies a number of competition “uncertainties” (as well as various consumer protection uncertainties) that the CMA plans to investigate going forward such as:

• Whether proprietary data, cutting edge performance, and important computational resources will be required to compete?
• Whether models will become large and generalized or specific and specialized?
• Whether “large technology companies” will have an advantage over others and if open-source models will remain a key part of the market?
• Whether companies and consumers will have the ability to make choices between models and whether switching models will be difficult?
• Whether “vertically integrated firms and partnerships have an incentive to foreclose upstream and downstream competitors?”

The CMA Report concludes by identifying the “guiding principles” essential to ensuring that “competition and consumer protection remains an effective driving force” as AI models develop. These include:

• Access: ongoing ready access to key inputs
• Diversity: sustained diversity of business models, including both open and closed
• Choice: sufficient choice for businesses
• Flexibility: flexibility to switch or use multiple models according to need
• Fair Dealing: no anti-competitive conduct, including “anti-competitive self-preferencing, tying or bundling”
• Transparency: giving information about the risks and limitations of model-generated content

The CMA plans to start a “significant programme of engagement” in the UK, U.S., and globally including speaking to a variety of stakeholders including “leading FM developers such as Google, Meta, OpenAI, Microsoft, NVIDIA, and Anthropic.”