Understand how AI tools are being used across your business, what information they can reach and where practical safeguards are needed.
Discuss your securityAI tools often arrive before anyone has decided how they should be used. The starting point is a plain picture of what is in use today.
What matters most is what goes in. Business plans and personal accounts are covered by different terms, and the controls on offer vary by product, plan and configuration, so the answer for your organisation depends on what you actually use.
What an AI tool can access depends on its connections, permissions and configuration. A review checks those arrangements and whether information is shared more widely than intended.
People need short guidance they can follow on an ordinary working day, and to know who to ask.
Microsoft's documentation for Copilot in Microsoft 365, the version organisations license for work, states that it only surfaces organisational data to which the individual user already has at least view permission. That makes your existing permissions and sharing the thing to check before it is switched on, or soon after. Other Copilot products and licences differ, so a review looks at the one you have.
A plain-English account of which AI tools are in use, how they are accessed and where information may be exposed.
The issues that matter most, in order, so effort goes where it reduces risk.
What to change and who will do it: us, your IT provider or your own team. Where we implement changes, you get written notes on what changed.
Organisations use different tools on different plans, so there is no fixed package. We start with a conversation about what you use and what concerns you. Together we agree which of the five areas to cover, which products are in scope and what you will receive. The work depends on your environment and available licences.
We name these products because they are widely used. What each one can reach, and how your information is treated, depends on the plan and how it is set up.
For Copilot in Microsoft 365, the version organisations license for work, Microsoft's documentation states that it draws on the emails, chats and documents the signed-in person has permission to access, and that prompts, responses and the data accessed are not used to train its foundation models.
Offered by OpenAI in plans for individuals and for organisations. OpenAI states that it does not train its models on business data by default. Its services for individuals have a data control that decides whether conversations are used.
Offered by Anthropic in plans for individuals and for organisations. Anthropic states that inputs and outputs from its commercial products are not used to train its models by default. On its consumer plans this depends on the choice made in privacy settings.
Checked against each company's published documentation on 10 October 2026. Terms and settings change, so a review checks the current position for the plans you use. Product names are trademarks of their respective owners and are used only to identify the products. Artificial Armour is independent and is not affiliated with, accredited by or endorsed by Microsoft, OpenAI or Anthropic.
Tell us which AI tools your business uses or is considering. We aim to respond within one business day.
