United Kingdom - AI Foundation Models Report

Competition and Markets Authority - Initial report on foundation models

United Kingdom

RAI-GB-NA-CMAIRXX-2023
In Force(In Force)
PolicyGovernance and OversightTransparency and DisclosureMarket Surveillance
Export PDF

The Competition and Markets Authority (CMA) published an initial review report on AI Foundation Models on 18 September 2023. The report maps the technical landscape, identifies competition and consumer protection risks, and sets out a set of principles and near-term actions the CMA will pursue to support competitive, transparent and safe markets for foundation models.

Summary

In September 2023 the Competition and Markets Authority (CMA) published an initial report on AI Foundation Models (FMs) following an initial review launched in May 2023. The report provides a comprehensive technical primer about how foundation models are developed (pre-training, fine-tuning, alignment), deployed (APIs, hosted services, open-source releases) and monetised, and assesses the likely competition and consumer protection implications of their rapid adoption. It reflects engagement with over 70 stakeholders and presents a spectrum of potential market outcomes — ranging from positive innovation-driven scenarios to concerning outcomes with entrenched market power, weak competition and consumer harms.

Key competition concerns identified include high fixed costs for pre-training, dependence on large datasets and compute, network and feedback effects, vertical integration risks where FM developers could foreclose downstream markets, and the potential for ‘winner-takes-most’ dynamics. On consumer protection, the report highlights risks such as false or misleading outputs (hallucinations), user manipulation, misleading advertising and the opacity of FM-generated outputs that can impede consumer understanding and redress.

Rather than prescribing new law, the CMA sets out principles to guide market development — covering contestability and access to inputs (data, compute), transparency and disclosure, accountability and governance, consumer protection safeguards, and competition policy vigilance (including merger control). The report explains how existing UK competition and consumer protection laws (including the Consumer Rights Act 2015 and Consumer Protection from Unfair Trading Regulations) apply to FM developers and deployers, and signals likely regulatory interest in mergers and conduct that could reduce contestability. It also describes the CMA’s next steps: continued monitoring, engagement with other UK regulators via the Digital Regulation Cooperation Forum, targeted market surveillance, and publishing updates (including an April 2024 update and technical update) while coordinating internationally.

For firms, the CMA’s findings mean increased regulatory scrutiny of business models that restrict access to critical inputs or degrade consumer protections. The report recommends adoption of transparency practices (clear disclosure when consumers interact with FM outputs), robust testing and evaluation, governance measures to limit harms, and consideration of contestability-enhancing remedies. While not a binding regulation, the report functions as an authoritative statement of CMA priorities for AI markets and a forward signal that existing competition and consumer protection powers will be actively used where risks materialise.

Full article

Read full text ↗

Overview

The Competition and Markets Authority (CMA) published its "AI Foundation Models: Initial Report" on 18 September 2023 after launching an initial review on 4 May 2023. The report provides an accessible technical overview of foundation models (FMs), their development lifecycle (pre-training, fine-tuning, alignment), and the commercial routes for deployment and monetisation. It maps potential benefits — innovation, productivity and new services — against competition and consumer protection risks such as high fixed costs and compute requirements, concentration risks from vertical integration, network effects and consumer harms like hallucinations or misleading outputs. The CMA frames the document as an initial review aimed at shaping early principles for FM markets and outlines next steps for monitoring, engagement and potential interventions. The full report and accompanying summary materials are published on GOV.UK; see the CMA report page AI Foundation Models: Initial report and the full report PDF AI Foundation Models: Initial Report (Full report PDF).

Definitions

The CMA defines foundation models (FMs) as large, pre-trained AI models trained on broad datasets that can be adapted for a wide range of downstream tasks and applications. Examples include large language models (LLMs) trained on text and multimodal models trained on images, audio or video. The report explains terminology such as pre-training (the stage where a model learns general patterns), fine-tuning (task- or domain-specific adaptation), inference (runtime model use), alignment (methods to make outputs safe and desirable), and deployment routes (hosts, APIs, open-source releases, licensing). The glossary clarifies concepts used throughout the report and situates FMs relative to other AI systems in both technical and economic terms.

Governance and Institutional Framework

The report situates the CMA’s work within the broader UK regulatory ecosystem. The CMA emphasises its distinct mandate for competition and consumer protection while recognising overlapping responsibilities with regulators focused on safety, data protection and sectoral rules. It notes engagement with the Digital Regulation Cooperation Forum (DRCF) and other bodies and highlights the potential impact of forthcoming UK legislative initiatives (e.g., the Digital Markets, Competition and Consumers Bill / Digital Markets Unit regime) on the CMA’s toolkit. The CMA states it will coordinate with other regulators to address safety, privacy and human-rights implications while focusing its remit on contestability, mergers, market conduct and consumer law enforcement. The CMA’s case page and timeline for the review are available on GOV.UK: AI Foundation Models: initial review (CMA case page).

Key Focus Areas

The report concentrates on three principal themes: (1) competition in the development of FMs, (2) impacts of FMs on competition in downstream markets, and (3) consumer protection. For competition in development the CMA analyses barriers to entry including data access, compute costs, technical expertise, funding and model evaluation benchmarks. It discusses open-source models as a counterweight to concentration but notes uncertainties about sustainability. For downstream markets the CMA examines vertical integration risks where dominant FM providers might foreclose rivals or privilege affiliated downstream products, data-feedback loops that amplify incumbents’ advantages, and switching costs for downstream firms. On consumer protection the report details risks from hallucinations, misleading outputs, advertising and disclosure failures, and insufficient redress routes. Across these areas the CMA proposes principles to promote contestability, transparency, accountability, safety, and consumer redress mechanisms to mitigate harms.

Implementation Framework

The report is not a prescriptive regulation but establishes a practical framework of principles the CMA will apply in its future work and interventions. These encompass ensuring open and contestable markets (access to data and compute), transparency and meaningful disclosure (clearly informing users when outputs are FM-generated and their limitations), testing and evaluation (robust technical testing to measure reliability and safety), governance and accountability (organisational controls, documentation and incident response), and active competition oversight (merger scrutiny, market studies and targeted enforcement). The CMA sets out how existing UK competition and consumer protection law applies to FMs — noting potential reliance on the Consumer Rights Act and Consumer Protection from Unfair Trading Regulations where relevant — and indicates how it will use its existing powers and forthcoming tools to support market outcomes consistent with the principles.

Monitoring and Evaluation

The CMA commits to an ongoing programme of monitoring and engagement. This includes gathering further evidence, tracking market developments (open-source releases, compute concentration, mergers and partnerships), stakeholder engagement (developers, downstream firms, consumer bodies), publishing updates and technical annexes, and conducting targeted market surveillance where risks crystallise. The CMA published an update paper and technical update in April 2024 to reflect market developments and stakeholder feedback. The approach emphasises iterative evidence collection and readiness to escalate to formal market investigations or enforcement where warranted.

Penalties, Liability, and Appeals

The initial report itself does not create new penalties but signals enforcement using existing legal frameworks. Potential enforcement instruments referenced include competition investigations (merger control, abuse of dominance), consumer protection enforcement under existing consumer law (including investigations, consumer remedies and civil actions), statutory powers available to the CMA and, where applicable, remedies such as undertakings, fines, orders or divestment in merger contexts. The CMA highlights that misrepresentations or unfair trading practices involving FMs could trigger action under consumer protection law. Where the CMA initiates investigations or decisions, statutory routes for appeal and legal review remain available to affected parties.

Relationship to Other Instruments

The report maps its analysis to existing UK statutes and regulatory instruments (competition law, consumer protection, data protection regimes), and situates the CMA’s activities relative to government AI policy (the UK AI White Paper) and other regulatory initiatives. It notes international developments such as the EU AI Act and other jurisdictions’ approaches, and affirms cooperation with other UK regulators through the DRCF to align responsibilities and avoid regulatory gaps or overlap. The CMA further outlines how sectoral regulators and international partners may be engaged where sector-specific risks (e.g., healthcare, finance) arise.

International Alignment

The CMA recognises the global nature of foundation models and the need for international coordination. The report references international regulatory developments and the value of information sharing, joint engagement with major jurisdictions, and coherence with multilateral efforts (including the UK government’s participation in international AI safety work). The CMA signals its intention to monitor and coordinate with counterparts to limit regulatory fragmentation while ensuring that UK competition and consumer protections remain effective in a global market.

Implementation Timeline

EventDate
Launch of CMA initial review2023-05-04
Report published (Initial report)2023-09-18
Stakeholder consultation deadline (initial phase)2023-06-02
Update paper published2024-04-11
Technical update report published2024-04-16

Sources and References

SourceType
AI Foundation Models: Initial Report (Full report PDF)Primary Source
AI Foundation Models: Initial report (GOV.UK page)Primary Source

Requirements for a company

What an organisation has to do under United Kingdom - AI Foundation Models Report, at a glance. Not legal advice — the table below gives the provision and deadline for each item.

Must do

5
  • Adopt clear user-facing disclosures when outputs are generated by foundation models.Providers and deployers of foundation models.
  • Provide clear routes for consumer complaints and remediation for foundation model outputs.Providers and deployers of foundation models.
  • Maintain records of foundation model testing, performance metrics, and mitigation strategies.Developers and providers of foundation models.
  • Assign accountability, establish incident response procedures, and maintain risk registers for foundation models.Developers and providers of foundation models.
  • Document access to data and compute, and steps to mitigate market lock-in.Providers of foundation models.

Must not do

0

Nothing in this category.

Should do

0

Nothing in this category.

Should not do

0

Nothing in this category.

Who must do what

The obligations under United Kingdom - AI Foundation Models Report, most serious first. Not legal advice — verify against the official text before relying on it.

#WhoRequirementBy whenWhereSeverity
1Providers and deployers of foundation models.Adopt clear user-facing disclosures when outputs are generated by foundation models.
transparency and meaningful disclosure (clearly informing users when outputs are FM-generated and their limitations)
Before placing on marketImplementation FrameworkCritical
2Providers and deployers of foundation models.Provide clear routes for consumer complaints and remediation for foundation model outputs.
consumer redress mechanisms to mitigate harms.
Key Focus AreasCritical
3Developers and providers of foundation models.Maintain records of foundation model testing, performance metrics, and mitigation strategies.
testing and evaluation (robust technical testing to measure reliability and safety)
Implementation FrameworkImportant
4Developers and providers of foundation models.Assign accountability, establish incident response procedures, and maintain risk registers for foundation models.
governance and accountability (organisational controls, documentation and incident response)
Implementation FrameworkImportant
5Providers of foundation models.Document access to data and compute, and steps to mitigate market lock-in.
ensuring open and contestable markets (access to data and compute)
Implementation FrameworkImportant

© Regulations.AI · updated on 13-Jun-2026