Summary

  • BEREC has opened consultation on a draft report that treats AI systems as a new layer of internet intermediation. It says the Open Internet Regulation governs internet-access services, while the selection and presentation power of AI systems operates in the content and application layer.
  • The draft points to a two-sided WebMCP effect: a common interface could make it easier to switch AI browsers, yet the platform controlling the user-facing agent could gain more influence over what users see and do. The proposal remains a community-group draft, not a W3C Standard.

The gateway has moved above the network

The Body of European Regulators for Electronic Communications opened comments on 7 October on its draft report about generative AI and internet openness. The Board of Regulators adopted the text at its 1–2 October meeting. Contributions are due by 11 November. The report is a consultation draft, not a new rule or a final finding about any company.

Its central move is to follow the user’s path beyond connectivity. A broadband or mobile provider carries traffic between a user and online services. Search and AI products increasingly decide which sources to retrieve, how to rank them, whether to answer with a summary, and when an agent should take an action. Both layers affect access, but they expose different control points.

BEREC’s draft says the EU Open Internet Regulation places equal-treatment duties on providers of internet-access services. The same law also protects end users’ ability to access and distribute information and use applications without discrimination; BEREC’s guidance recognises businesses and content or application providers when they use internet access. That protection still matters when an ISP blocks or slows traffic. It does not, the draft argues, give regulators an operative tool over an AI platform’s own selection and presentation of content.

BEREC says the Regulation’s aims remain a useful reference, while other instruments may address parts of the higher-layer problem.

That distinction keeps two questions apart. Is a network provider treating traffic unlawfully? And is an AI service shaping discovery or action through its interface, ranking, defaults or source selection? A finding about the second does not establish a breach of the first. Nor does BEREC’s draft say that AI intermediation sits outside every European law: it examines the AI Act, Digital Markets Act, Digital Services Act and Data Act, and says coordination may matter because no single instrument covers the full set of concerns.

A browser protocol could redraw the handoff

The report’s WebMCP discussion makes the control shift concrete. The W3C Web Machine Learning Community Group’s document, dated 9 October, describes an API through which a website can expose JavaScript tools to AI agents. In the report’s example, a shopping site can offer tools to filter products or update a cart inside the browser session. The agent’s platform supplies the reasoning and user interface; the site supplies callable functions.

This could make an agent more reliable than asking it to interpret every page from screenshots or rendered code. It could also reduce the need for a person to navigate each site manually. BEREC therefore sees two plausible effects. A shared interface may make it easier for rival AI platforms or browsers to reproduce website functions, lowering switching costs. At the same time, if the AI platform remains the user’s main screen, it can retain control over discovery, filtering and commercial presentation while the website becomes a service endpoint behind it.

Neither effect is established. WebMCP is a Draft Community Group Report, not a W3C Standard, and BEREC says the incentives for website operators to adopt it remain an open question. The report describes a possible architecture, not a measured change in traffic or a finding about deployment. Its value as evidence is that the shift can be specified: who chooses the tool, orders the choices, controls the interface, sees the user’s intent, and receives the human visit or only an automated request?

Access can persist while discovery changes

For publishers, creators and public-interest collections, the relevant outcome is not just whether a page remains reachable. AI answers can summarise material without sending a user to its source; agents can interact with a site without a conventional visit. BEREC warns that this may weaken referral traffic, advertising or subscription opportunities even as AI products continue to depend on online material. Its report also notes that automated retrieval can create additional operating costs for websites. These are risks and mechanisms identified in a draft review, not a quantified forecast for every publisher.

The distinction matters to telecom operators too. Network availability, traffic management, interconnection and application-layer discovery are related but separate records. A successful reachability test cannot show that a source was visible in an AI answer. A fall in site referrals does not by itself show that an ISP discriminated against the traffic. Operators, platforms and content providers need evidence at the layer where a decision was made.

Comments to BEREC can now test whether the draft has mapped those interfaces correctly: how AI services select and cite sources; whether users can see, challenge or change those choices; whether common protocols actually reduce switching barriers; and how websites can measure automated access and resulting costs. The published consultation page asks contributors to identify the report sections they address. BEREC says it will publish contributions and a summary, while respecting clearly identified confidentiality requests.

The test is adoption, choice and observable effect

Heng Lu’s “Running-Code Primacy” supplies a useful editorial test: a proposal should not be mistaken for a deployed result. The same caution applies to regulation. A draft’s account of a gap does not decide which authority should close it, whether existing powers suffice, or how a remedy would affect users and providers.

For the final report, the evidence to watch is practical. Does BEREC distinguish network traffic equality from interface-level selection? Does it identify measurable duties without prescribing a single AI interface? Does WebMCP gain adoption across browsers and sites, and can users switch agents without losing useful site functions? Do referral, source-diversity and bot-cost measures show a change that can be attributed to AI mediation rather than ordinary search or market shifts?

The policy stakes are not that every answer engine is a telecom network. They are that a user can retain a technical path to the web while losing visibility into how that path is curated. The consultation is an opportunity to define where network neutrality ends, which instruments govern the next layer, and what evidence would show whether either users’ choice or publishers’ ability to reach them has changed.

Sources