AI Tools and Cyber Essentials: Are ChatGPT, Copilot and AI SaaS Services in Scope?
Published 16 October 2026
AI SaaS is a cloud service like any other
NCSC Requirements for IT Infrastructure v3.3 describe three cloud service types: infrastructure (IaaS), platform (PaaS) and software (SaaS). If your organisation uses a cloud service, it is in scope, and you keep responsibility for the configuration and access controls you manage, even though the provider runs the platform.
A web-based AI assistant, an AI writing tool, an AI meeting note-taker or a coding assistant are all, in Cyber Essentials terms, SaaS. The "AI" label does not change the category. Our general guide to Cyber Essentials and cloud services explains the shared-responsibility model in more detail.
Work accounts vs personal accounts
| How the tool is used | Cyber Essentials position |
|---|---|
| Organisation-provided business or enterprise account | In scope as a cloud service. Apply MFA, user and admin controls. |
| AI built into a service you already use (for example Copilot in Microsoft 365) | Covered through that existing service's identities and settings. |
| Staff sign up personally using their work email address | Treat as organisational use. Either bring it under a managed account or stop it. |
| Staff use a personal account on a personal device, for personal purposes | Not an organisational cloud service. But if organisational data goes in, it becomes a governance problem. |
The awkward middle case is "shadow AI": staff using free accounts for work because nobody gave them a sanctioned tool. The cleanest Cyber Essentials answer is to provide an approved tool on managed accounts and say clearly which tools are allowed.
Accounts, MFA and admin ownership
- MFA: authentication to cloud services must use MFA. Ideally sign in through your identity provider (single sign-on) so your existing MFA policy applies. If the tool has its own logins, enable and enforce its MFA.
- User accounts: individual named accounts, created through a joiner process and removed when someone leaves.
- Admin accounts: know who administers the AI tool's workspace or tenant, keep that to a small number of people, and protect admin sign-in with MFA.
- Ownership: one person or team should own each AI tool, so someone is accountable for its settings.
The devices used to reach AI tools
The devices staff use to access in-scope cloud services are relevant too. A company laptop running an AI desktop app, or a phone with an AI app signed into a work account, must meet the applicable Cyber Essentials controls for its device type: for example a supported operating system, CE-required vulnerability fixes within 14 days, and the secure configuration and malware protection requirements that apply to that kind of endpoint. Personally owned devices that access organisational data or services are in scope as well, which is why BYOD rules matter here. See Cyber Essentials and BYOD.
What to record in your cloud-service inventory
- Tool name and provider.
- Plan or licence type (business, enterprise, built into another service).
- How users sign in (single sign-on or separate accounts) and how MFA is enforced.
- Named admin(s) and owner.
- Which teams use it and on which devices.
- Any connectors or integrations that give it access to other systems.
Examples
A business ChatGPT workspace: list it as a SaaS service, connect sign-in to your identity provider if supported, enforce MFA, and restrict workspace admin rights. OpenAI's own documentation explains which sign-in and SSO options apply to each plan.
Copilot in Microsoft 365: it sits inside your Microsoft 365 tenant, so the Cyber Essentials work is the tenant itself: Entra ID accounts, MFA, admin roles and the devices staff use. We cover this separately in Microsoft 365 Copilot and Cyber Essentials.
A niche AI tool one team signed up for: the most common gap. Find it, decide whether it is approved, and if so bring it under the same account and MFA rules.
What Cyber Essentials does not assess
Cyber Essentials is a technical baseline for preventing common internet attacks. It does not assess:
- How the AI model was trained or how accurate it is.
- Whether staff write good prompts or check outputs.
- What data you allow into AI tools, retention settings or training opt-outs.
- Data protection compliance, contracts or supplier due diligence beyond the CE controls.
Those questions still matter, but they belong to AI governance and privacy, not Cyber Essentials. Our AI governance framework for SMEs covers that layer. If you are an AI company yourself rather than a user of AI tools, see Cyber Essentials for startups and AI companies.