Every company now wants the same thing: an AI that knows everything the company knows. Every deal, every email thread, every document, every number — one brain to ask instead of fifteen apps to search. The technology for this exists. What stops most companies is not capability. It is trust.
An AI that can retrieve across everything will leak across everything. Ask it an innocent question and it helpfully surfaces the salary sheet, the board deck, the acquisition that half the company isn't supposed to know about. The more complete the brain, the worse the exposure — which is why so many "AI knowledge" rollouts quietly stall at a demo over public data.
The fix is not a smarter filter bolted onto the AI. It is an architecture in which permission is a property of the knowledge itself. In Atollon, everything lives in a business context — a client, a project, an initiative. Access rights are set on the context, and everything stored inside inherits them automatically: emails, documents, tasks, calendar items, financial transactions. Not permissions configured per feature, per module, per integration — one access model that every object obeys.
When AI answers a question, it inherits the context rights of the person asking. It can reason across everything that person is entitled to see — and is structurally unable to surface anything else. Same question, different person, different answer. That is not a restriction on the brain; it is what makes the brain deployable.
For AI to act on your company's knowledge, it has to respect who is allowed to know what. Context access rights are the guardrail that makes a Company Brain safe to turn on.
Here is the part we could not fake if we started today: Atollon has been building context-based access control with inheritance since 2000. It was not added because AI made the leak risk obvious — it has been the platform's architecture for a quarter of a century, refined against real organisations the entire time. Tools that secured one data type and are now retrofitting permissions around an AI feature are discovering what we learned early: a permission model only holds if everything obeys it, and that cannot credibly be bolted on. It has to be the foundation.
The Company Brain that knows everything — and shows each person only what they're meant to see. That is the standard we think every company should hold its AI to. We happen to have spent 26 years building the architecture that meets it.