Most pages about secure AI are written by a vendor selling one answer. This guide sets out the questions a company has to settle, from training terms to who approves what AI does, and what each option commits to in writing, with every claim traced to a page a reader can open.
Who it is for
Owners, chief executives, finance directors and heads of IT at established companies, typically from 20 to a few thousand people, whose staff already use AI assistants and who want that use governed. It is also for companies about to give AI access to their mail, files or systems for the first time.
How it is made
Each option is described from its vendor's own pages: pricing pages, product pages, privacy documentation and terms. Independent research, such as IBM's breach study and the OWASP list of AI application risks, is cited with its publisher and date. Statements a vendor makes about its own service are attributed to it. Where a vendor's page did not say something, we write that it was not stated or not assessed. The order of entries and the criteria are explained on the methodology page.
Use of product names
ChatGPT, Claude, Microsoft 365 Copilot, Gemini, Glean, Azure OpenAI, Amazon Bedrock and the other names in this guide identify the products discussed and nothing more. We use no vendor logos or brand colours.
Updates and corrections
Vendors change their data terms and prices often. The guide is reviewed every quarter and sooner when a listed vendor changes a term. Each page shows its last update and a change log. If anything here is wrong or out of date, write to us.
