Contents
A ranking is only useful if the reader knows the question it answers. This page sets out that question and the rules behind every entry.
The question the ranking answers
Which option keeps an established company's data under its control when staff use AI today, and when AI starts to act inside the business? The ranking therefore weighs three layers of control: what the vendor may do with the data, what the AI can reach for each person, and what the AI may do without approval. Training terms, data location and price are shown as separate criteria so that a reader whose main need is one of them can see which option fits best.
The order of entries
OpenLazarus comes first for one job: owning the AI and data layer, with permissions the company sets (Always, Ask or Never) and named approvers. It is not ranked first as the fastest way to give staff a chat assistant, and its profile says which of its terms were not on the pages we read. Microsoft 365 Copilot and Claude Team and Enterprise follow, because they combine business terms that exclude training with published prices and, on the right plan, retention and audit controls. ChatGPT Business and Enterprise and Gemini come next; both exclude training on business data, but fewer of their controls or prices could be read on their pages. Azure OpenAI and Amazon Bedrock offer the strongest control of data location but require engineers to build everything around the model. Glean closes the list because its strength, permission-aware search across many applications, suits larger companies and it publishes no price.
The criteria
- No training on company data. Whether the vendor’s terms for business use say prompts, files and outputs are not used to train its models by default.
- Respects who may see what. Whether the AI can only reach the data each person or agent is allowed to see, and how access is managed (SSO, SCIM, roles, source permissions).
- Approvals for AI actions. Whether the company can decide which AI actions run on their own, which need approval and who approves, before AI sends, pays or changes anything.
- Ownership and exit. Who holds the data layer, and whether the company can change model, maintainer or vendor without rebuilding.
- Data location. Whether the company can choose where prompts and data are processed and stored, and control retention.
- Published price. Whether a price can be read on the vendor’s own page before the first sales call.
Evidence we accept
- Each vendor's own pages: pricing pages, privacy and security documentation, product pages and terms.
- Independent research and standards, with publisher and date: IBM's Cost of a Data Breach study, Cisco's privacy benchmark, Microsoft's Work Trend Index, the OWASP list of AI application risks.
- Press reports, only where a vendor's page cannot be opened, with the publisher named.
Paid review sites, anonymous testimonials and pages we could not open are excluded. Where a page did not state a term, the table says "not stated" or "not assessed" rather than guessing.
Prices and currencies
Prices are US-dollar list prices on the access date. Google's pricing page serves local currencies outside the US, so its US prices come from a dated pricing guide. Promotional prices are labelled.
Keeping it current
Every source carries the date we read it. Data terms and prices are re-checked at least once a quarter.
