Debate on AI safety

What is AI safety?

AI safety is about making sure AI systems operate in ways that benefit humans, and are aligned with human values, rather than causing harm or negative outcomes to society.  The field of AI safety is focused on designing AI systems that are reliable, secure, robust and behave as intended. The aim is to prevent, or at the very least minimise, risks from AI including:

• accidents (unintended harms arising from AI systems failing, or acting in unexpected ways);
• misuse (including generating disinformation and malicious external threats, like cyberattacks, data poisoning and data breaches) and;
• bias (unfair or discriminatory outcomes).

What measures are used to put AI safety into practice?

There is not a single approach to ensuring that an AI system operates safely. The Alan Turing Institute, the UK’s national institute for data science and artificial intelligence, has emphasised the importance of considering safety measures at every stage of the lifecycle of a project involving AI, from design through to development and deployment. It adds that the exact safety risks faced will depend on factors like the sort of AI models being used, how those AI models are applied and deployed, where the training data has come from and how the objective of the AI system has been specified.

Examples of safety measures include:

• Model selection and training: ensuring that the task at hand is most appropriately addressed by AI and, if it is, that the AI model selected meets the needs of the project.
• Data extraction and analysis: checking training data for its accuracy, reliability and quality. This includes “considering and evaluating whether the data is of appropriate relevance, timely, complete and representative, and in sufficient quantity to meet needs of the AI system”.
• Testing and validation: testing an AI system before its deployed to check it behaves consistently and does not produce unwanted outcomes.
• System monitoring: regularly reevaluating an AI system, post-deployment, to check its performance and reliability and to ensure it is keeping pace with real world changes.

Other methods for mitigating risks include AI auditing (external, third party scrutiny of an AI system), transparency registers (collecting data on AI systems that are currently in use, like the UK’s Algorithmic Transparency Recording Standard for public sector bodies) and ethical frameworks / committees to oversee and direct the use of AI.

How is AI regulated in the UK?

The UK does not have any AI-specific regulation or legislation covering the technology. Instead, AI is regulated in the context in which it is used, through existing legal frameworks, and is overseen by existing regulators covering specific sectors, such as Ofcom (the UK’s communications regulator), Ofgem (the energy regulator in Great Britain), and the Financial Conduct Authority (the UK’s conduct regulator for financial services).

The UK also relies on:  

• Non-statutory principles for AI governance (see, for example, Department for Science, Innovation and Technology, Implementing the UK’s AI Regulatory Principles, February 2024, PDF)
• Targeted legislation like the Online Safety Act 2023. The 2023 act was introduced to regulate online platforms and combat illegal and harmful content. Under the act, Ofcom is empowered to enforce compliance with regards to online services which includes AI chatbots. The government has published an Online Safety Act: explainer (April 2025).

The Conservative Government established the AI Safety Institute (AISI, since renamed the AI Security Institute) in 2023.  The AISI says it is “conducting research and building infrastructure to understand the capabilities and impacts of advanced AI and to develop and test risk mitigations”. The government has also published reports on AI safety and held safety summits.