Last reviewed: 16 November 2026. Dates below are taken from the legal texts linked. This is not legal advice.
What changed
The EU AI Act, Regulation (EU) 2024/1689, entered into force on 1 August 2024 and applies in stages:
- 2 February 2025: prohibited practices and AI literacy duties.
- 2 August 2025: obligations for providers of general-purpose AI models.
- 2 August 2026: the general date of application for most remaining provisions.
- 2 December 2027: obligations for high-risk systems listed in Annex III. This was originally 2 August 2026 and was moved by Regulation (EU) 2026/1744.
- 2 August 2028: obligations for high-risk systems in products covered by Annex I, moved from 2 August 2027 by the same regulation.
The UK has taken a different route. Its 2023 white paper, A pro-innovation approach to AI regulation, asks existing regulators to apply five principles within their own remits. Check for later UK legislation before relying on this section.
The EU AI Act and UK businesses
Under Article 2, the Act can apply to a UK business that:
- places an AI system or general-purpose AI model on the EU market, or
- provides or deploys an AI system whose output is used in the EU.
The Act sorts AI systems by risk.
Prohibited practices (Article 5)
- Social scoring that leads to unjustified or disproportionate treatment
- Real-time remote biometric identification in public spaces for law enforcement, with narrow exceptions
- Emotion recognition in workplaces and education, except for medical or safety reasons
- Manipulative techniques, and exploiting vulnerabilities such as age or disability, in ways that cause significant harm
High risk (Annex III and Annex I)
- Critical infrastructure such as energy, transport and water
- Education, including admissions and grading
- Employment, including recruitment, promotion and dismissal
- Access to essential services, including credit scoring and some insurance pricing
- Law enforcement and border control
High-risk systems need a risk management system, data governance, technical documentation, human oversight, and measures for accuracy and robustness.
Transparency duties (Article 50)
- Chatbots must tell people they are talking to an AI system
- Deepfakes must be labelled
- Where emotion recognition is permitted, the people exposed to it must be told
Minimal risk
Most other systems, such as spam filters and recommendation engines, have no specific obligations under the Act.
The UK approach
The 2023 white paper sets five cross-sector principles:
- Safety, security and robustness
- Appropriate transparency and explainability
- Fairness
- Accountability and governance
- Contestability and redress
Each regulator applies them in its own area: data protection for the Information Commission, financial services for the FCA, communications for Ofcom and competition for the CMA.
Building an AI governance framework
Step 1: List your AI systems
- Which AI systems do you use, including third-party tools?
- What does each one do?
- What data does it process?
- Who is affected by its outputs?
- What happens if it fails or produces biased outputs?
Step 2: Classify risk
For each system, assess:
- Effect on individuals. Does it make or influence decisions that materially affect people?
- Autonomy. Does a human review its outputs before they take effect?
- Data sensitivity. Does it process special category data?
- Scale. How many people are affected?
Step 3: Write the policies
- Accountability. Who is responsible for each system?
- Transparency. How do you explain its decisions to the people affected?
- Fairness. How do you test for bias and reduce it?
- Human oversight. When and how do people review outputs?
- Monitoring. How do you track performance over time?
Step 4: Put controls in place
- Carry out DPIAs for AI systems that process personal data in high-risk ways
- Write a model card for each system: purpose, training data, limitations
- Set testing procedures for bias and accuracy
- Define what happens when a system fails
- Monitor for model drift
Step 5: Train your team
Developers, product managers and business leaders need to know:
- When and how to escalate an AI-related risk
- How to document design decisions
- How to test for bias
- When a DPIA is required
Practical recommendations
- Start with your highest-risk system. You do not need to govern everything at once.
- Use an existing framework. ISO/IEC 42001 is the international standard for AI management systems.
- Keep AI governance and data protection distinct. They overlap, but AI governance also covers fairness and accountability where no personal data is involved.
- Write down your decisions. A record of what you considered and why is the main evidence of governance.
- Review regularly. AI systems change over time.
Our AI Governance Framework service helps UK businesses set this up.
Sources
- Regulation (EU) 2024/1689 (the AI Act), Articles 2, 5, 6, 50 and 113, and Annexes I and III.
- Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), which amended the application dates in Article 113.
- Department for Science, Innovation and Technology, A pro-innovation approach to AI regulation, March 2023.
- ISO, ISO/IEC 42001:2023, AI management systems.
