Is ChatGPT Safe for Client Data? A Plain Guide for Law Firms

Short answer: the free version of ChatGPT is not a safe place for client data. Business versions of ChatGPT, Claude and Microsoft Copilot can be used safely, but only with the right plan, the right contract and clear rules for your staff.

Most partners we speak to already suspect someone in the firm is pasting client material into a free AI tool. This guide explains the real risks, what the SRA now expects, how the different ChatGPT plans compare, and the practical steps a firm can take this month.

Why the free version is a problem

Free and personal AI accounts are built for consumers, not regulated businesses. The main issues are:

• Training on your data. On Free and Plus, chats can be used to improve the model unless each user switches this off themselves. Most people never do.
• No contract with the firm. A fee earner's personal account sits outside your data protection agreements. There is no data processing agreement between the firm and the provider.
• No oversight. You cannot see what was shared, by whom, or delete it centrally. When someone leaves, their chat history leaves with them.
• Confidentiality and privilege. Putting privileged or personal data into an unapproved tool can breach your duty of confidentiality and UK GDPR. In UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), the Upper Tribunal said that putting client letters into an open AI tool such as ChatGPT places them in the public domain, breaching client confidentiality and waiving legal privilege. The tool itself is not the risk. The lack of control is. For a deeper look at where your data actually goes, read our guide: What happens to your client's data when you use an AI tool?

What the SRA says

In August 2026 the SRA published a formal warning notice on the use of AI. It does not ban AI or create new rules. It reminds solicitors that their existing duties still apply, whatever tool helped produce the work.

The SRA's two main concerns are made up case law and client confidentiality. It received 42 reports of potential AI misuse between July 2025 and July 2026, many involving fake citations reaching the courts. On confidentiality, its position is that client information should only go into an AI system when proper contractual, technical and organisational safeguards are in place.

The notice also makes clear that supervisors share responsibility. If a trainee's AI drafted submission reaches a court with false citations, the supervising solicitor may be in breach too.

In short, AI does not change your duties. It changes how you meet them.

ChatGPT plans compared

OpenAI offers several plans, and the differences matter for law firms.

• Free and Plus. Personal plans. Training on your chats is on by default unless each user opts out. No admin controls and no firm level contract. Not suitable for client data.
• Business. This was called ChatGPT Team until August 2025. It is priced from around $25 per user per month on annual billing, with a minimum of two seats. Business data is not used for training by default, and it includes a shared workspace, admin controls and single sign on.
• Enterprise. Custom pricing for larger organisations, with stronger security, compliance and support options. For most small and mid size firms, Business is the realistic starting point. Prices and features change often, so always check the current terms before you sign up.

What safe use looks like

Choosing the right plan is only the first step. A safe setup usually includes:

• A signed data processing agreement with the provider, reviewed by whoever handles data protection in the firm.
• Single sign on, so access is tied to firm accounts and ends when someone leaves.
• Admin controls switched on, including who can use connectors to email, drives and other apps.
• Custom GPTs kept inside the workspace. Do not publish them publicly, as outside users may be on plans with different data rules.
• Retention settings agreed, so you know how long chats and files are kept.
• A data protection impact assessment before you start using the tool for personal data. The same principles apply to Claude Team and Microsoft 365 Copilot. Each has different strengths, and the right choice often depends on the systems you already use.

Five steps for your firm

1. Find out what is already happening. Ask staff, without blame, which AI tools they use and for what. You will almost certainly find personal accounts in use. This is shadow AI, and it is far more common than most partners think.
2. Pick one approved tool. Choose a business plan and switch off anything you do not need.
3. Write a short AI policy. One or two pages on what can and cannot go into the tool, and how outputs are checked. Keep it readable. You can start from our free AI policy template.
4. Train your people. A policy nobody understands changes nothing. Partners and fee earners need hands on practice with real, safe examples. Our guide to AI training for solicitors explains what that looks like.
5. Review every quarter. The tools change fast, and so does the guidance. Your rules should keep up.