Analysis Cybersecurity AI Compliance

Shadow AI and UK GDPR: your exposure and what to do after a leak

Why staff using personal AI accounts create exposure under UK GDPR, what the NIS Regulations and ISO 27001 expect, and how to respond when client or personal data has already gone into an unapproved AI tool.

10 November 2026
Updated 18 November 2026
4 min read

This page covers the legal and compliance side of shadow AI. For what shadow AI looks like and the three risks that matter most, see our shadow AI guide. Where it hides and how to build a response staff will follow are in finding and controlling shadow AI.

Why UK GDPR exposure is larger than most teams realise

An AI provider may act as your processor, handling personal data on your behalf, or as an independent controller that decides for itself how it uses the data. Which role applies depends on the service, the contract and the settings, and it determines which obligations follow. Article 28 is the UK GDPR provision most often overlooked in shadow AI discussions. Where a vendor processes personal data on your behalf, it requires a written contract setting out the subject matter, duration, nature and purpose of the processing, the type of personal data and categories of data subjects, and your obligations and rights as controller.

A member of staff logging into ChatGPT with a personal account has no such contract with you. You do not know the processing is happening. No purpose limitation, safeguards or deletion schedule have been agreed. Your organisation is the controller of the personal data that employee handles in their job, and sending it to an external AI service is processing, whatever the employee intended.

Key point

Where an AI provider acts as your processor, UK GDPR Article 28 requires a written contract. A personal AI account gives you no contract and no say over how the provider uses the data, and the provider may act as a controller in its own right. Your organisation remains accountable for the personal data it controls and for having appropriate security measures. Whether a specific disclosure leads to enforcement or a claim depends on the facts.

Not every use of an unapproved tool is a reportable breach. A prompt that contains no personal data raises confidentiality questions, not UK GDPR ones. Where personal data has been sent to an external AI tool without authorisation, treat it as a possible personal data breach and assess the facts. If it is a breach, you must record it, and report it to the ICO within 72 hours of becoming aware if it is likely to result in a risk to people's rights and freedoms. Without visibility of AI use, you cannot make that assessment at all.

The ICO has published guidance on AI and data protection, and supervisory authorities in Italy, Germany and Spain have already opened investigations or taken action over AI tools and personal data. The Irish Data Protection Commission, which leads on many large US technology companies operating in the EU, has AI firmly on its agenda.

What NIS rules and ISO 27001 require

UK GDPR Article 32 already requires security appropriate to the risk, and that includes knowing where personal data goes. For network and information systems, UK operators of essential services and relevant digital service providers fall under the NIS Regulations 2018, and organisations providing services in the EU may be in scope of NIS2. Both expect in-scope organisations to manage risks from third-party services, which can include AI services handling sensitive data. If you are not in scope, they are still a useful benchmark.

ISO 27001:2022 is risk-based. The Annex A controls most relevant to AI tools are 5.9 (inventory of information and other associated assets), 5.10 (acceptable use of information and other associated assets) and 5.23 (information security for use of cloud services). The standard does not require a standalone AI policy; you can cover AI in your existing acceptable use and cloud policies. It does require you to identify the risk, decide how to treat it and keep evidence.

If you hold or are pursuing ISO 27001 certification, expect auditors to ask how AI tool use is reflected in your risk assessment, asset inventory and acceptable use rules. Where AI use is widespread and none of these address it, an auditor may raise a nonconformity; how serious it is depends on the risk and the evidence you can show.

What to do when data has already gone into an unapproved AI tool

Sooner or later someone will tell you, or your monitoring will show, that client or personal data went into a personal AI account. Treat it as a possible incident and work through it in order.

  1. Establish the facts: which tool and account, what data, how many people or records, when, and whether the account is shared or synced to other devices.
  2. Preserve evidence before anything is deleted: a record of what was shared, screenshots of the conversation and any relevant logs. Collect only what you need and respect the employee's own privacy.
  3. Contain it: once evidence is preserved, ask the employee to delete the conversation and, where the provider allows, clear the history, unless a legal hold or the investigation requires the data to be kept. Switching off model training prevents future use but does not delete what has already been sent. Check the provider's actual retention, because deletion may not take effect straight away and copies may be kept for a period.
  4. Assess the risk to the individuals: the type and sensitivity of the data, what the provider's terms allow it to do with the data, and whether it could be misused.
  5. Decide on notification: if personal data is involved and the breach is likely to result in a risk to individuals, report it to the ICO within 72 hours of becoming aware, and tell the individuals if the risk is high. Record the breach and your reasoning either way.
  6. Check other obligations: client contracts, NDAs, sector regulators and insurers may have their own notification terms.
  7. Fix the cause: offer an approved alternative, update the policy or guidance, and respond to the employee fairly. Staff who are punished for reporting a mistake are less likely to report the next one.

Frequently asked questions

Does using a personal ChatGPT account at work violate GDPR?

It depends on what is entered and on what terms. If no personal data is involved, UK GDPR may not apply, although confidentiality obligations still can. If a staff member pastes personal data about customers, patients or colleagues into a personal AI account, your organisation has disclosed that data to a provider it has not assessed and has no contract with, often outside the UK. That can breach UK GDPR principles on lawfulness, security and international transfers, and it may be a personal data breach that you must record and, where there is a risk to individuals, report to the ICO within 72 hours. Responsibility sits with your organisation as controller; the outcome depends on the data and on how you respond.

Ryland Deakin
About the author
Lead Consultant, Cyvra · CISM · CompTIA Security+ · MCP

Ryland has delivered cybersecurity, compliance, and IT management programmes for regulated organisations across the UK and the Netherlands for over 20 years. View full profile

Talk to Cyvra

Map your shadow AI exposure

We help you identify which AI tools your staff are using, build a policy they will follow, and deploy enterprise tools that give them a compliant alternative.

Disclaimer: This article is for general informational purposes only and does not constitute legal, regulatory, or professional advice. Cyvra makes no warranty as to the accuracy or completeness of this content. Readers should seek independent advice appropriate to their specific circumstances. Cyvra accepts no liability for any loss arising from reliance on this content.