Meta under the microscope: could Meta’s recommendation algorithms be prohibited under the EU AI Act?

Estimated reading time: 6 minutes

On 26 August 2026, a US federal court entered a consent judgment resolving claims by state attorneys general concerning Facebook and Instagram. The proceedings were brought as a result of a series of whistleblower disclosures by Meta executives and employees beginning in 2021. The states alleged that Meta’s product design, including its recommendation algorithms, encouraged compulsive use by children and teenagers and harmed their mental and physical wellbeing. Meta denied the allegations. Although the claims were resolved without the court deciding whether the allegations were true or Meta had breached the law, the settlement of up to $18 billion had a significant impact on financial markets.

In the first instalment of our three-part series on the Meta case, we consider whether Meta’s recommendation technology qualifies as AI, whether the EU AI Act could prohibit its use and how the UK approach compares.

The next two articles in this series will consider the whistleblowing lessons for UK employers, and the impact of the settlement on investors and financial markets.

What was alleged?

The states’ complaint alleged that Meta’s “Recommendation Algorithms” used signals including likes, follows and lingering on content to choose or arrange what each user saw. It also alleged that Meta designed those systems “to maximize youth engagement” through unpredictable content sequences and engagement targeted at individual users.

When is a recommendation algorithm AI?

Article 3 of the EU AI Act defines an AI system by how it operates, not by how its supplier labels it. It must be machine-based, operate with some autonomy and infer from inputs how to generate outputs, expressly including recommendations, which can influence physical or virtual environments.

The alleged features correspond to several parts of that definition. Behavioural signals may be inputs, engagement the objective and selected or arranged content the output. What the allegations do not establish is whether the system inferred how to generate those outputs.

The Commission’s guidance confirms that both AI and non-AI systems can generate recommendations. An AI recommender may rely on machine learning or logic-based inference. A non-AI system may simply execute fixed human-programmed instructions without “learning, reasoning or modelling”.

The alleged use of behavioural signals to personalise content selection is therefore consistent with an AI-based recommender, but is not conclusive. Classification would require technical evidence about the particular system’s architecture and functionality.

Could Article 5 prohibit the system?

If the recommender qualifies as AI, Article 5(1)(b) prohibits its use where it exploits a vulnerability arising from age, disability or a specific social or economic situation, with the objective or effect of materially distorting behaviour in a manner that causes or is reasonably likely to cause significant harm. The Commission’s non-binding guidance says there must be a “plausible causal link” between the exploitation, distortion and harm.

The guidance gives an example of how those requirements may be met. It describes a game which analyses children’s behaviour and provides personalised, unpredictable rewards through reinforcement schedules to encourage excessive and compulsive use. The guidance says this exploits vulnerabilities including children’s limited understanding of long-term consequences and lack of self-control.

The similarities to the allegations against Meta do not establish that Article 5 would apply. Increased engagement does not necessarily amount to material distortion. Evidence would still be needed that the system exploited an age-related vulnerability and that the resulting distortion caused, or was reasonably likely to cause, significant harm.

The DSA applies without AI classification

The EU Digital Services Act regulates recommender systems whether or not they qualify as AI. It defines them as fully or partly automated systems used by online platforms to suggest or prioritise information. Platforms accessible to minors must take appropriate and proportionate measures to ensure a high level of privacy, safety and security. Very large platforms and search engines must assess and mitigate risks from their design and algorithms, including risks to minors and mental wellbeing, and offer a non-profiled option for each recommender system they use.

In July 2026, the European Commission preliminarily found Meta in breach of the DSA over the addictive design of Facebook and Instagram. It referred to highly personalised recommenders, infinite scroll, autoplay and push notifications. Meta can respond before the Commission reaches a final decision.

How does UK law address the same risks?

UK law addresses the alleged risks through overlapping online-safety and data-protection rules without first classifying the recommender as AI. There is no statutory definition of AI that applies across UK law, nor an AI-specific equivalent to Article 5’s age-vulnerability prohibition.

Under the Online Safety Act, in-scope services likely to be accessed by children must assess how algorithms and features affect children’s exposure to harmful content and take proportionate measures. That assessment must be repeated before a significant change to the service’s design or operation. Ofcom’s guidance identifies recommender systems and features affecting how much children use a service as risk factors.

Since 5 February 2026, Article 25 of the UK GDPR has required controllers of in-scope online services likely to be accessed by children to take specified child-protection matters into account when deciding what safeguards to put in place. These include how best to protect and support children, their lower awareness of data-processing risks and their differing needs by age and development.

The Children’s Code is the ICO’s statutory code of practice for online services likely to be accessed by children. It helps explain how Article 25 and other UK GDPR duties apply in practice. Where a recommender system uses a child’s personal data to profile their interests or behaviour, the Code expects features that rely on profiling to be off by default unless the provider can demonstrate a compelling reason, taking account of the child’s best interests. It also calls for safeguards against harmful recommendations and says data-driven features that make it difficult to disengage are likely to breach the fairness principle.

After the US outcome, the ICO referred to the Children’s Code and said it was contacting Meta, while noting its ongoing work examining platforms’ use of children’s personal information in recommender systems.

Could the UK introduce dedicated AI legislation?

Parliamentary pressure for dedicated AI legislation is growing. In a report published on 14 September 2026, the Joint Committee on Human Rights compared the UK with the EU AI Act and concluded that “There are no equivalent prohibitions in the UK”. It recommended a dedicated, risk-based AI Bill that would prohibit uses inconsistent with human rights, with the detail of those prohibitions settled after public consultation.

These are parliamentary recommendations, not current law or government policy. Asked about the report the following day, the Government maintained its “agile and context-based approach” and said it would respond fully “in due course”. It is expected to respond within two months but has not committed to legislation. No government AI Bill or consultation on such a Bill has been published.

Key takeaways

  1. Classify the technology, not the marketing. An “algorithm” is not automatically AI. Document how a recommender generates rankings and whether it infers from inputs.
  2. Treat children’s risk as a design and compliance For UK services in scope, document how algorithms and engagement features affect different age groups, carry out a further risk assessment before significant design or operational changes, and apply proportionate measures. Where recommendations involve profiling, review profiling defaults and safeguards against harmful content.
  3. Apply each regime separately. A recommender that is not an AI system under the EU AI Act may still be subject to the DSA or UK law. Meet current obligations while monitoring UK reform.

Businesses developing, deploying or procuring AI systems and AI-enabled products or services should review the classification and governance of those technologies under UK and EU law. Please contact Kolvin Stone, James Collis or another member of our Technology team to discuss the regulatory and commercial implications for your business.

Portfolio Close
Portfolio list
NameCVEmail
No one added to your portfolio yet.

Remove All

Download

Close

Exclusive Content

Unlock 24-hour access to our exclusive content including on-demand webinars by submitting your details using the form below:

"*" indicates required fields

Name*
Agreement*