Summary

  • Ofcom’s September review sketches agents that could compare broadband offers and prepare a move, while leaving the binding choice with the customer.
  • The commercial hinge is proof of authority: who may act, within what scope, and what evidence reaches the provider before a contract becomes effective.

The most revealing part of Ofcom’s AI review is a transaction that has not happened yet. A personal assistant monitors a household’s broadband, compares price, speed, contract length and switching steps, fills in the paperwork, then waits for the customer’s consent. Ofcom labels the scenario as future-facing. Its significance is the handoff: an agent can remove search and form-filling work, but a recommendation is not authority to bind a customer.

That distinction is easy to blur when an interface compresses research, comparison and purchase into a single conversation. A user may grant an assistant permission to inspect bills and service quality. That permission does not necessarily authorize it to choose among trade-offs, accept a contract summary, transfer personal details or commit the household to a new minimum term. Each step has a different consequence and may require a different authorization.

Ofcom’s September 25 paper updates work first opened in January. It finds that customer-facing AI in telecoms is developing cautiously. The quantitative evidence is a useful scale check, not proof of agent adoption: Ofcom’s 2026 Technology Tracker interviewed 4,033 UK adults aged 16 and over between March and June. Forty-eight per cent said they had used generative AI in the previous year; 23 per cent had interacted with customer-service chatbots. Eight per cent said they had used an AI service in relation to telecoms, including six per cent for an issue, three per cent for comparison and two per cent for a complaint.

These uses overlap, and the questions measure reported interaction rather than successful switching.

The report’s qualitative work points to a less tidy customer journey. It combined 35 interviews and 13 focus groups with consumers, small-business decision-makers and people with additional needs. Participants valued speed and convenience, but their preferences differed: some wanted a person for a serious problem or final negotiation, while others preferred not to speak to one. The defensible lesson is choice at the moment it matters, not a universal preference for either channel.

Ofcom already has a contract rule to apply. General Condition C1.6 says a contract becomes effective only after the relevant customer has received the contract summary and given express consent to enter it. Ofcom’s paper says providers should consider how that requirement works when a customer has authorized an AI agent to prepare a deal or purchase a service. It does not declare that every agent workflow meets the rule, nor does it create a new agent-specific consent regime.

The distinction shifts attention from how clever the recommendation sounds to the transaction record. A reliable handoff would need to show which customer was authenticated; whether the agent could browse, recommend, negotiate or execute; which service and provider the permission covered; what contract summary the customer received; and when the customer approved the final terms. If price, speed and term are presented as a composite “best deal”, the customer must still be able to see the features the agent ranked and the compromises it made.

An agent paid by a provider or comparison platform could otherwise optimise for a legible score while concealing a commercially shaped shortlist.

Identity verification is only one part of this control surface. Ofcom says common cross-platform protocols that would let people sign up or switch in third-party environments are not generally available today. It also identifies fraud, agent manipulation and sensitive-data exposure as future risks. A signed credential could help a provider verify that a permission came from the account holder, but it cannot by itself prove that the permission was informed, current, narrow enough or based on an accurate offer comparison.

The regulatory boundary is also specific. Ofcom said there was limited evidence that AI applications had caused systemic harm to telecom customers and that existing rules currently protect them. It clarified that customers should know when a GenAI application is involved, and warned against AI becoming a gatekeeper that blocks access to support channels providers are required to make available. This is not a blanket right to a human for every exchange. It does mean that an automated front door cannot quietly turn a required route to support into an inaccessible maze.

The near-term business problem is therefore not an abstract contest between human agents and machines. Providers and intermediaries must decide where assistance ends and delegated action begins. Keeping the customer’s approval step visible may make the journey slower than a fully autonomous agent, but it also preserves a traceable point at which offer, terms and authority meet. Until those records and authentication paths work across providers, a promising comparison tool is not yet a dependable switching agent.

Sources