Guides / Is AI safe for visa applications?

Is it safe to use AI for UK visa applications?

It can be, if the tool is set up for the data it handles and your firm stays in control of every application. This guide covers what to check: special category data under UK GDPR, who is controller and processor, how long documents are kept, what the regulators say, and why an AI tool should never submit.

Last updated: 15 October 2026

This guide is general information for firms, not legal advice. Your firm's own data protection position depends on its work and should be confirmed with your DPO or adviser.

Why immigration files need extra care

Under UK GDPR, Article 9 sets out "special category" data: racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic data, biometric data used to identify someone, health, sex life and sexual orientation. The ICO notes that inferences can count too (ICO).

A typical immigration file can contain several of these. TB test results and medical evidence are health data. Relationship evidence can reveal sexual orientation. Passports, birth certificates and supporting letters can reveal ethnic origin or religion. Criminal record information is not special category data, but it has its own protections under Article 10 of UK GDPR and the Data Protection Act 2018.

Processing special category data is prohibited unless you have both an Article 6 lawful basis and an Article 9 condition, plus a Schedule 1 condition of the Data Protection Act 2018 where one is required (ICO).

The two conditions most often discussed for client work are explicit consent, Article 9(2)(a), and Article 9(2)(f): processing "necessary for the establishment, exercise or defence of legal claims or whenever courts or tribunals are acting in their judicial capacity" (UK GDPR Article 9). Whether preparing an immigration application falls within legal claims is a legal judgement for your firm and its adviser, and the ICO notes its guidance is under review following the Data (Use and Access) Act 2025. Using an AI tool does not change which condition you rely on, but your privacy notice to clients should cover it.

Controller and processor

For most firms, the firm is the controller of its clients' data: it decides why and how the data is used. A software vendor that handles the data only on the firm's instructions is a processor (ICO).

That means you need a written data processing agreement with the vendor, and you need to know which sub-processors it uses and where they are. A vendor that uses your clients' data for its own purposes, such as training its models, is acting as a controller for that use, and you would need a basis for sharing it.

What the regulators say

The SRA's warning notice on the misuse of AI, published in August 2026, says:

• AI has no legal personality. Solicitors remain responsible for work done with it, and must be competent to use it and supervise it.
• Client information should only go into AI systems with appropriate contractual, technical and organisational safeguards. Data should be kept secure, not used to train models unless explicitly authorised, and not kept longer than necessary.
• Putting confidential information into public AI tools is likely to breach confidentiality and may waive privilege.

The SRA's compliance tips on AI and technology add that firms should have governance, risk assessments, policies and training in place, and should be clear with clients when AI is used.

As of 15 October 2026 the Immigration Advice Authority (formerly the OISC) has not published guidance specifically on AI. Its Code of Standards and its guidance on competence still apply to any adviser using AI tools.

What to check before you upload a client's file

1. Is there a data processing agreement? It should limit the vendor to your instructions and list its sub-processors.
2. Where is data stored and processed? Ask about every provider that sees client documents, including the AI model provider, and what safeguards cover any transfer outside the UK.
3. How long are documents kept? The shorter the better. Ask what happens to originals, extracted data, screenshots and logs.
4. Is client data used for training? The answer should be no, in writing.
5. Who reviews the output? A qualified person at your firm should check every extraction and every completed form.
6. Can it submit? It should not be able to.
7. Is there a DPIA? Processing special category data with new technology is likely to call for one. Ask to see the vendor's and do your own.

Document deletion

The safest copy of a passport is the one that no longer exists. A form-filling tool needs the documents only long enough to read the details out of them. After that, the extracted summary is what the firm reviews and what fills the form.

Look for a tool that deletes originals as soon as they have been read, deletes the rest of the working data shortly after the form is filled, and has a backstop for runs that fail halfway. Your firm keeps its own file in its own systems, as it does now.

Why AI should never submit

The declaration at the end of a Home Office application confirms that the information given is true and complete. That is a professional judgement, and it should never be delegated to software.

AI tools make mistakes: a misread date, a name in the wrong order, a missed trip abroad. Those are catchable when a caseworker reviews the completed form before submission. They are much harder to fix once an application has been submitted and the fee paid. A tool that stops before submission keeps your firm's review as the final control.

How Aion handles this

• Your firm is the controller, Aion is the processor, under the data processing agreement in Schedule 1 of our Terms. Before each application your firm confirms its lawful basis, including the Article 9 condition, and that confirmation is logged.
• Documents are deleted on extraction. Originals are removed as soon as the details have been read out. Everything else about an application is deleted automatically within 24 hours, with a storage-level deletion rule as a backstop.
• Each application runs in its own single-use container in AWS London, which is destroyed when the work ends. Its browser can only visit gov.uk.
• No training on client data. We use Anthropic's Claude through its commercial API, under terms where Anthropic does not train its models on our data. AI processing takes place in the United States under the UK's approved transfer safeguards.
• Two human checkpoints. Your caseworker reviews the extracted summary before anything is filled, and reviews the completed application before it is submitted.
• Aion never submits. It stops at the Declaration page with a report of anything missing or uncertain. Your team submits.
• Honest gaps. Aion does not yet hold SOC 2 or ISO 27001 certification. We will complete your security questionnaire and share our full DPIA on request.

The full detail, including every sub-processor and the DPIA summary, is in our security and data protection summary. How we handle personal data about you and your firm is in our privacy notice.

Sources: ICO: What is special category data? · ICO: Rules on special category data · ICO: Controllers and processors · SRA: Misuse of AI warning notice · SRA: Compliance tips on AI and technology.