ChatGPT security risks for businesses

Is ChatGPT a security risk for businesses?

ChatGPT itself is a mature service from a large provider. The risk for most organisations is not ChatGPT as such; it is how it is used:

• Staff using personal accounts (Free, Plus or Pro) for work, where OpenAI's consumer terms apply rather than its business commitments
• Sensitive data pasted or uploaded without anyone checking: client files, personal data, credentials, source code
• Connected apps and agents that let ChatGPT read and act on other systems
• No record of what was shared, by whom, if a client or the ICO asks

Approving a business plan solves some of this, not all of it. This guide covers both, using OpenAI's own published material. We do not repeat news stories or speculate about incidents; the risks below follow from how the service works.

Personal accounts vs ChatGPT Business and Enterprise

OpenAI's enterprise privacy commitments cover ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers and the API. Its services for individuals work differently. From OpenAI's own pages:

For personal accounts, OpenAI says users can opt out of training by turning off Improve the model for everyone in ChatGPT's data controls or through its Privacy Portal, and that temporary chats are not used for training. It also notes that on individual plans, giving thumbs up or thumbs down feedback can bring the whole conversation into training even after opting out. None of those choices is under the organisation's control, which is the core problem with personal accounts at work.

ChatGPT Enterprise security controls

For organisations that approve ChatGPT, these are the security controls OpenAI describes for its business plans. Some apply to every business plan and some only to Enterprise (and Edu and Healthcare), so check the plan you are buying:

• No training on business data by default: OpenAI says it does not use prompts, responses or files from ChatGPT Business, Enterprise or Edu to improve its models unless the organisation has signed a separate agreement allowing it, and that thumbs up or down feedback on those plans does not permit training on the conversation
• SAML single sign-on: enterprise-level authentication through your identity provider
• Encryption: AES-256 at rest and TLS 1.2 or higher in transit
• SOC 2 Type 2: completed for ChatGPT Business and ChatGPT Enterprise
• Retention control: workspace admins control how long data is kept on ChatGPT Business and on Enterprise, Edu and Healthcare; OpenAI says deleted conversations are removed from its systems within 30 days unless it is legally required to keep them (for Business, also where longer retention is reasonably necessary to protect its services or others from harm)
• Audit log (Enterprise, Edu, Healthcare): workspace admins can access a log of conversations and GPTs through the Enterprise Compliance API
• Business admin access: on ChatGPT Business, workspace admins can view, access, export and delete end users' conversations in the workspace
• Apps: workspace admins control which apps are enabled, and OpenAI says it does not train its models on data accessed from apps by default
• Ownership: OpenAI says you own your inputs and outputs, where allowed by law
• Data processing addendum: OpenAI can sign a DPA for ChatGPT Business, Enterprise and the API in support of GDPR compliance, which matters for UK GDPR accountability

These are solid foundations, and for many organisations an approved ChatGPT plan is a sensible part of an AI strategy. Detailed plan features change, so check OpenAI's current documentation when you buy.

The ChatGPT security risks an enterprise plan does not solve

An enterprise plan protects data that goes into the enterprise workspace. It does not control what happens outside it:

• Personal accounts on the same service. The business and personal versions of ChatGPT share a web address. A user signed in to a personal account looks much the same to a network filter.
• Every other AI tool. Approving ChatGPT does not stop people pasting the same data into other chatbots such as Gemini, Claude or DeepSeek, AI extensions or AI features in SaaS apps. See what is shadow AI.
• Data that should not be there at all. Even in an approved workspace, some data should not be shared: special category data, client material under strict confidentiality, credentials. A business plan has no view on what your policy allows.
• Connected apps widen the blast radius. Each app or connector lets ChatGPT read from, and sometimes act on, another system. Over-broad permissions there become visible quickly.
• Prompt injection. When ChatGPT browses the web or reads documents and connected apps, hidden instructions in that content can try to steer it. See prompt injection.
• Output risk. AI output can be wrong or inappropriate and still be sent to a client. That is a policy and review issue.

Closing these gaps is about policy plus controls that sit where people actually use AI: in the browser and on the device.

A ChatGPT security policy: what to include

Most organisations do not need a separate ChatGPT policy; it fits inside an AI acceptable use policy. The ChatGPT-specific points to cover are:

• Which plan is approved, and that work must use the company workspace, not a personal account
• Which data types may and may not be entered or uploaded, even in the approved workspace
• Who may enable apps and connectors, build GPTs, or share GPTs outside the organisation
• That AI output must be reviewed by a person before it is relied on or sent to a client
• What is monitored, and that some data may be redacted or blocked
• How to report data shared in error

Under UK GDPR, personal data entered into ChatGPT is still your organisation's responsibility as controller. Our guide to AI and UK GDPR and our AI DPIA guide cover the data protection side. This is general information, not legal advice.

How Workforce AI Security closes the gaps

Check Point Workforce AI Security sits in the browser and, in the Enterprise tier, on the device, which is where the remaining ChatGPT risks live. Check Point names ChatGPT among the public AI tools it covers. Based on Check Point's material, it can:

• Discover who uses ChatGPT and which other AI apps, extensions and agents are in use, with session-level visibility
• Apply different policies to managed and unmanaged apps, so the approved workspace and personal use can be treated differently
• Redact personal data, credentials, source code and financial data from prompts and uploads inline, with labelled placeholders, including text inside files and images using OCR
• Restrict copy and paste and file uploads where needed
• Block unauthorised AI apps, and guide users to the approved tool with an explanation
• Record what happened in audit trails and reports

If people use ChatGPT in the browser, Workforce AI Security Essentials covers web-based AI apps. If they use the ChatGPT desktop app, coding tools or agents, that is desktop and agentic AI use, which is what Workforce AI Security Enterprise adds. We confirm how specific apps and account types are handled in a pilot rather than assume it. For the bigger picture, see AI data leakage.

How we help

Forever Systems can help you get ChatGPT working safely: deciding whether an approved plan makes sense, writing the policy, and using Workforce AI Security to see personal-account use, redact sensitive data and steer people to the approved workspace. We start with a discovery pilot so decisions are based on your real usage. Ask us for a quote.

To talk about ChatGPT security in your organisation, contact us.