Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThe UK does not currently have one universal, legally binding AI-cybersecurity rule for every organisation. Its central instrument is the UK AI Cyber Security Code of Practice, published by the Department for Science, Innovation and Technology (DSIT) on 31 January 2025. The code is explicitly voluntary, although it sets detailed baseline expectations for organisations that develop, deploy, operate or supply AI systems.
Those expectations sit alongside binding data-protection, sector, contractual and other cybersecurity duties that may apply to your organisation. NCSC guidance, the Cyber Governance Code of Practice and the published ETSI technical specification add practical and technical context, but they do not turn the voluntary UK code into a statute.
As an Amazon Associate I earn from qualifying purchases.
Which UK AI security rules apply in October 2026?
Think of the current framework as several layers rather than a single law. The legal force and intended audience of each layer are different.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11| Instrument | Current status | What it is for | Who should use it |
|---|---|---|---|
| UK AI Cyber Security Code of Practice | Voluntary; published 31 January 2025 | Baseline cybersecurity expectations across an AI system’s lifecycle | Developers, system operators, data custodians and organisations using AI |
| AI Cyber Security Code implementation guidance | Government support material | Practical detail for applying the code’s provisions | Teams implementing or assessing the code |
| NCSC Guidelines for Secure AI System Development | Guidance version 1.0, published and reviewed 27 November 2023 | Secure design, development, deployment, operation and maintenance | AI providers, developers, managers, risk owners and decision-makers |
| Cyber Governance Code of Practice | Board and director framework | Ownership, risk appetite, assurance, culture, strategy and training | Public- and private-sector boards, especially medium and large organisations |
| ETSI TS 104 223 | Technical specification version 1.1.1, April 2025; related EN transposition publication recorded 12 December 2025 | Baseline technical requirements for securing AI models and systems | Organisations seeking a technical baseline or standardisation reference |
| Data-protection and sector requirements | Potentially binding, depending on circumstances | Rules for personal data, regulated activities, contracts and sector risks | Each organisation must assess its own legal and regulatory position |
The code uses mandatory-sounding words such as “shall” inside a document that identifies itself as voluntary. That wording describes the code’s expectations; it does not by itself create a statutory duty for every UK organisation.
#1 Best Overall
What systems and organisations are in scope?
The code is aimed at AI systems intended for deployment. It covers systems using deep neural networks, generative AI and other machine-learning approaches. An academic system created and tested only for research, with no intended deployment, is outside the code’s stated target. Once a research project is prepared for operational use, its security obligations should be reassessed.
Responsibility follows activities, not simply the name on a supplier contract. The code identifies four roles:
| Role | Responsibility in practice | Typical questions |
|---|---|---|
| Developer | Creates or adapts an AI model or system | What threats were considered? How were the model, prompts, data and software tested? |
| System operator | Embeds or deploys the model or system in organisational infrastructure; this can include a provider embedding AI for business customers | Which environment, interfaces, permissions and monitoring are under our control? |
| Data custodian | Controls permissions and integrity for data used by the system and sets data-use or management policies | Is the data appropriate, accurate, authorised and protected for this use? |
| End user | Employees and UK consumers who use the AI system | What warnings, safeguards, reporting channels and training do users receive? |
One organisation can occupy several roles. For example, a company that fine-tunes a model and deploys it in its own customer-service platform is both a developer and a system operator. Several data custodians may share responsibility for different datasets. Buying an external model does not remove the need for risk assessment, supplier due diligence or clear internal accountability.
Recommended Free Tools
What the voluntary code expects across the AI lifecycle
The code groups expectations into five phases. The following is a practical reading of those phases, not a substitute for the full code and its implementation guide.
1. Secure design
Before building or buying an AI capability, developers and system operators should document the business need, security risks and proposed mitigations. The assessment should explain why AI is appropriate, what could go wrong, who could be harmed and which controls reduce the risk.
Designers should threat-model AI-specific attacks as well as ordinary application threats. Examples named in the code include data poisoning, model inversion and membership inference. The model should also be considered in the context in which it will operate: connected applications, retrieval stores, plug-ins, identity systems and business processes can create attack paths that are not visible in an isolated model.
2. Secure development
Developers should protect the integrity of source code, training and tuning data, model artifacts, configuration, prompts and build pipelines. The code calls for an audit trail covering operation and lifecycle management of models, datasets and prompts. Keep records sufficiently detailed to reconstruct what changed, who approved it and which version was deployed.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
When external components, hosted models or APIs are used, developers and operators should perform AI-security risk assessment and due diligence within existing software-assurance processes. A supplier’s security claims are evidence to assess, not a transfer of your responsibility.
3. Secure deployment
Deployment controls should be proportionate to the system’s function and potential impact. Systems should be designed and implemented to withstand adversarial AI attacks, unexpected inputs and AI system failure. That is an explicit statement in the government code.
Where an AI system can call other systems or data sources, assess every permission and grant only what the required functionality needs. Separate development, testing and production environments where practical; protect service accounts; authenticate API calls; and constrain tools that can write data, send messages, execute code or change records. Document the assumptions behind those controls.
4. Secure maintenance and operation
NCSC guidance treats security as a continuing requirement rather than a launch gate. Monitor model behaviour, access, prompts, connected services and infrastructure for signs of misuse or compromise. Logging should support detection and investigation while respecting data-protection requirements.
Maintain an incident process that covers AI-specific events such as prompt injection, poisoned data, unauthorised model extraction, leakage of sensitive information and unsafe or misleading output that affects a business process. Define escalation routes, decision-makers, customer communications and recovery procedures before an incident occurs.
Updates can change the threat profile. Revisit threat models and approvals when the model, system prompt, retrieval source, connected tool, data or configuration changes. Test rollback or safe-disable procedures, and keep a record of which version produced an output involved in an incident.
5. Secure end of life
Plan how models, datasets, prompts, credentials, logs and integrations will be retired. Revoke access, remove obsolete endpoints, preserve records needed for legal or security purposes, and dispose of sensitive artifacts according to applicable information-management rules. A system that is no longer advertised can remain exposed if an API, scheduled job or service account is still active.
A practical implementation plan
The code is voluntary, but using it as an internal control framework gives teams a repeatable way to show what they considered and why.
Rank #4
- Inventory AI use. List models, applications, agents, plug-ins, datasets, prompts, APIs and business processes. Include unsanctioned tools discovered through procurement, expense or network reviews.
- Assign roles. For each system, name the developer, operator, data custodian and business owner. Record where one organisation performs more than one role and where a supplier or customer performs another.
- Document the business and risk assessment. State the intended use, unacceptable uses, affected people, assets, confidentiality, integrity and availability consequences, and the mitigations required before launch.
- Threat-model the complete system. Include data poisoning, model inversion, membership inference, prompt injection, supply-chain compromise, credential theft, denial of service and unsafe interaction with connected tools. Repeat the exercise after material changes.
- Classify and protect data. Match the sensitivity of training, tuning, retrieval and operational data to access controls, retention, encryption and monitoring. Personal data, including pseudonymised data, may trigger data-protection obligations; consult the relevant Information Commissioner’s Office guidance and your data-protection lead.
- Assess suppliers and components. Check model provenance, update practices, incident notification, logging, data-use terms, subcontractors, support boundaries and exit arrangements. Require evidence that external providers address the code’s expectations where that is appropriate to the relationship.
- Constrain permissions and interfaces. Use least privilege for users, services, tools and data sources. Apply allow-lists, network boundaries, rate limits and human approval for high-impact actions.
- Test before release. Test expected and adversarial inputs, failure modes, access controls, logging, resilience and recovery. Confirm that users understand limitations and know how to report a suspected security issue.
- Operate and monitor. Review logs and alerts, investigate anomalies, measure drift and check that safeguards remain effective. Keep an audit trail for model, dataset, prompt and configuration changes.
- Exercise incidents and retirement. Run a serious-incident scenario, identify the accountable decision-maker, test containment and rollback, and maintain a documented end-of-life process.
What boards and senior leaders should ask
The Cyber Governance Code of Practice places AI security within wider board responsibilities: risk management, strategy, culture, assurance and training. It is described as foundational in DSIT’s modular cyber-code approach. Organisations implementing the AI Cyber Security Code should also consider the governance code and any organisation-specific codes that apply to them.
NCSC’s leadership material asks two useful questions: “Do you understand where accountability and responsibility for AI/ML security sit in your organisation?” and “How would you respond to a serious security incident involving an ML tool?” A board should be able to answer both without relying on an informal understanding between the technology and legal teams.
The governance code reports general cyber figures from the Cyber Security Breaches Survey 2024: 50% of businesses, 66% of high-income charities, 70% of medium businesses and 74% of large businesses reported some form of cyber breach or attack in the previous 12 months. These are not AI-incident or AI-adoption rates; they illustrate the broader cyber environment in which AI systems operate.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How data protection and other laws fit
The voluntary AI code does not replace the UK data-protection regime, employment duties, confidentiality obligations, sector regulation, contractual controls or incident-reporting requirements. If an AI system processes personal data, including pseudonymised information, determine the applicable legal basis, transparency, security, retention, rights and international-transfer requirements with competent advice.
Regulated organisations should map the AI system to their existing obligations rather than assume that adopting the code creates compliance. A healthcare, financial-services, public-sector or critical-infrastructure deployment may face requirements that are not described in this general code. Contracts can also make a voluntary standard effectively mandatory between the parties.
Best Value
What ETSI TS 104 223 changes—and what it does not
ETSI TS 104 223, Securing Artificial Intelligence (SAI); Baseline Cyber Security Requirements for AI Models and Systems, version 1.1.1 is dated April 2025. The ETSI work-programme entry records publication of the related EN transposition on 12 December 2025.
This gives organisations a current technical reference for baseline requirements. It does not, on the evidence available here, make every provision a statutory duty for every UK organisation, and publication of a standard does not convert the UK government’s voluntary code into law. Check procurement terms, regulator rules and sector-specific requirements to establish whether a particular standard is required in your situation.
Common mistakes and how to correct them
- Calling the code “UK AI law”. Describe it as voluntary and identify the separate laws that may apply to the deployment.
- Assigning all responsibility to the model vendor. Keep an internal owner, document the operator’s controls and perform supplier due diligence.
- Securing only the model. Extend threat modelling to data, prompts, retrieval, APIs, tools, identity, infrastructure and business processes.
- Skipping change reviews. Reassess risk after model, prompt, data, connector or configuration changes.
- Ignoring users. Provide training, reporting channels, usage boundaries and safeguards for employees and consumers.
- Keeping no evidence. Maintain decisions, approvals, test results, audit trails, incidents and retirement records.
- Assuming pseudonymisation removes regulation. Treat pseudonymised data as potentially personal data and obtain a specific assessment.
- Leaving abandoned interfaces live. Retire endpoints, credentials, scheduled jobs and connected services, not just the visible user interface.
Bottom line
As of 13 October 2026, the UK’s main AI-specific cybersecurity instrument is a detailed but voluntary code. Use it as a lifecycle control framework, combine it with NCSC secure-development guidance and board-level cyber governance, and separately identify every binding data-protection, sector, contractual and regulatory duty that applies to your organisation. The practical test is whether you can show who is accountable, what risks were assessed, which controls protect the complete AI system, how incidents will be handled and how the system will be retired.
Frequently Asked Questions
Is the UK AI Cyber Security Code of Practice legally mandatory?
No. The code published on 31 January 2025 is explicitly voluntary. Other laws, regulator rules or contracts may still impose binding duties on a particular organisation or deployment.
Does buying an AI model transfer security responsibility to the supplier?
No. The organisation deploying or embedding the model remains responsible for assessing its own use, permissions, integrations, data and operational risks, while conducting appropriate supplier due diligence.
Are pseudonymised datasets outside UK data-protection requirements?
Not automatically. The code notes that personal data, including pseudonymised data, may bring data-protection obligations. The specific status depends on the data and circumstances.
What is the status of ETSI TS 104 223?
Version 1.1.1 is a published ETSI technical specification dated April 2025, with the related EN transposition recorded as published on 12 December 2025. It is a technical baseline, not evidence of a universal statutory requirement for every UK organisation.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
