EU AI Act expected to cover products with built-in AI in Northern Ireland from 10 November

Firms that make, import or distribute physical products with built-in artificial intelligence for sale in Northern Ireland are expected to have to meet parts of the EU AI Act from Tuesday 10 November 2026, once a decision of the UK-EU Withdrawal Agreement Joint Committee adds the Act to the Windsor Framework. The Department for Business, Innovation, Science and Trade set out how the arrangement would work in guidance published on Wednesday.

The Act will not apply in Northern Ireland in full. It covers AI only where the system is closely integrated with a tangible product placed on the Northern Ireland market. Software that people download or reach through a browser, an app or an API is outside it, as is the way businesses use AI tools in their own work. The direct duties the Act places on providers of general purpose AI models do not apply there either.

According to the guidance, the arrangement follows an exchange of views the UK requested in the UK-EU Withdrawal Agreement Joint Committee on 29 April 2025, which concluded on 7 May 2026, with the decision to add the Act to the Windsor Framework due at the committee's meeting on 10 November. The agenda for that meeting, published by the Cabinet Office on Wednesday, lists an update on the Windsor Framework and decisions to be adopted.

The closely integrated test

A product is in scope if the AI system is embedded or pre-installed when the product goes on sale, or if it is downloaded separately but the product is designed to rely on that one system for its advertised function.

The guidance works through several cases:

• a smart speaker sold with a built-in AI voice assistant is in scope
• a smartwatch whose advertised health monitoring depends on an AI app supplied by the manufacturer and downloaded during setup is also in scope
• a chatbot a customer chooses to download onto a laptop after buying it is treated as standalone software and is out
• a chatbot that the laptop maker pre-installs before sale brings the laptop into scope, though only for transparency duties because a chatbot is not a high-risk system

The same manufacturer making both the product and the AI, or the AI only working with that product, does not by itself make the two closely integrated. Manufacturers are told to keep records of major software or firmware updates, since these can change whether a product falls inside the rules.

Bans and deadlines

From 10 November, products containing a closely integrated AI system that uses one of the Act's banned practices cannot be placed on the market. Two of the bans apply later, from 2 December 2026. The guidance's examples for Northern Ireland include a connected toy that pushes children towards increasingly risky activities for praise, an augmented reality headset built around addictive reward loops aimed at younger users, and an AI camera that detects employees' emotions. Firms with such a product already on sale should expect to withdraw and recall it, disable the feature where possible, tell their supply chain and report to the relevant market surveillance authority. The ban covers placing products on the market, not the use of them.

Two other duties start on the same date. Providers must take steps to build AI literacy among staff and others who operate the systems for them, and products that talk to people must make clear the user is dealing with AI where that would not be obvious. The guidance contrasts a robot vacuum cleaner that answers voice commands, which is plainly AI-driven, with a humanlike voice assistant in a speaker, which may not be.

From 2 December 2026, products that generate synthetic audio, images, video or text through a closely integrated system must mark that output in a machine-readable way so it can be detected as AI-generated. A connected toy that produces AI speech is one example given.

High-risk rules follow later. Systems in the Act's Annex III use cases, such as a road traffic monitoring camera or a warehouse wearable that scores workers' productivity, are covered from 2 December 2027, but only when built into a physical product. AI acting as a safety component in products such as toys, lifts, medical devices and gas appliances, where the product needs third-party conformity assessment, is covered from 2 August 2028.

Deployers of high-risk systems in Northern Ireland have one limited duty: to monitor the system and report risks and serious incidents to the provider and the market surveillance authority. The guidance says European Commission guidance remains the main technical and legal reference for meeting the obligations.