Security
Hand over your company’s data with peace of mind.
OS iXanagi holds what matters most to a business — products, customers, money, and what your people know. This page explains, without jargon, how we protect it, and what we cannot protect.
Six promises
What we promise, first.
Each one is how the system is built — not a promise to be careful.
Your data never trains AI
There is no step in this system that trains AI on your data. It is used only to understand it and make it searchable.
Other companies cannot see it
Each company has its own section. Brands within one company are kept apart too. Knowledge every brand uses is placed at the company level.
Important actions wait for a person
Sending, publishing, money and deletion happen only with your approval, or within a scope, budget and time limit you set when delegating to the AI. Knowledge, too, is adopted by you as a rule. The exceptions are items whose basis is confirmed in the source material, and items you chose to adopt in bulk at import — in both cases the reason is recorded.
What the AI did is on record
Who asked, what the AI did, who approved it — in the same record as people’s actions, so you can trace it later.
What people can do depends on their role
Able to edit but not publish, no access to amounts, only their own brand. You set it to fit your company.
What we cannot protect, we say so
Our development rules forbid claiming protection we do not have. The second half of this page lists what we do not cover.
How it is enforced
Not rules on paper. Protection built into the structure.
Every vendor says “we never train AI on your data”. The difference is whether that is a line in a contract, or the way the system is built.
| Protected by rules | OS iXanagi | |
|---|---|---|
| Where the check happens | Buttons are hidden on screen. A request that skips the screen may not be stopped. | Checked where the work actually lands. With or without the screen, it stops the same way. |
| Whose request it is | Trusts the company name written in the request. Change it, and you can claim to be another company. | Decided from who signed in and where they belong. Rewriting the request does not make you another company. |
| Actions by the AI | The AI has its own way in, and records and permissions may work differently. | Screen buttons and the AI use the same way in. The AI cannot exceed the permissions of whoever asked, and it is recorded in the same place. |
| When it cannot decide | Lets it through, just in case. | Does not let it through. It stops and records that it could not decide. |
| Newly added features | If someone forgets the protection, it ships without it. | Requiring sign-in is the default. Even if someone forgets, it cannot be used without signing in. |
Your data
How we handle what you entrust to us.
What it is used for, where it is kept, and how it is kept apart.
To erase your company’s data in one go, please contact us. We confirm the scope first, then carry it out, and keep a record of what was erased and how much.
Common attacks
Well-known tricks are stopped at the door.
Web services are attacked in well-known ways. For each, there is a set way to stop it.
People stay in charge
The AI proposes. People decide.
The AI does a lot of work, but anything that goes out of the company passes a person’s approval or a scope a person delegated. Company knowledge, too, is adopted by a person as a rule.
The AI proposes
Article drafts, explanations that need updating, “the way we do things” it found. Each arrives first as a proposal.
A person reads and decides
Approve, edit or reject. Reject, and it stops there.
Only approved or delegated actions go out
Sending, publishing, money and deletion happen only with your approval, or within a scope, budget and time limit you set when delegating to the AI. Drafts and internal edits are made by the AI, and the previous version is kept.
It is recorded
Who asked, what the AI did, who approved. You can follow it later in a list.
Roles and permissions
Who can do what is set by role.
Common roles are ready from the start. Below are the defaults; the owner can change them to fit your company.
| What they can do | OwnerOwns the account | AdminRuns day-to-day work | MemberWrites and edits |
|---|---|---|---|
| Write and edit pages and articles | Yes | Yes | Yes |
| Publish | Yes | Yes | No |
| Change the site’s structure and design | Yes | Yes | No |
| Bulk import and roll back | Yes | Yes | No |
| See business numbers such as sales and profit | Yes | Yes | No |
| See who was paid how much (numbers about people) | Yes | No | No |
| Create your own roles | Yes | No | No |
Only their own brands
You can limit which brands each person can reach. Other brands’ rows do not even appear in their lists.
Just this person, just this range
“Only admins publish, but this person may publish for this one brand.” You can hand over a part without promoting them.
You cannot give what you do not have
Even whoever creates roles cannot hand out permissions they do not hold. The system closes that back door to promotion.
Keys for outside tools: narrow and short-lived
Keys for tools such as browser extensions can be limited in what they do, which brands they reach, and when they expire. If the holder’s role is lowered, every key they issued weakens at once.
How we verify
We try to break it ourselves.
Reading the code and thinking “looks fine” is not verification. We actually attempt the attack and record whether it was stopped.
Attack it
Call it directly without the screen. Claim to be another company. Try to raise your own role. Call it without signing in. We actually do it.
Record the result
Not “the test passed”, but “this attack was stopped with this response”. The steps and anything to watch for are kept too.
Turn it into an automatic check
Once an attack is stopped, it is checked automatically before every release. A change that removes a protection is stopped before shipping.
Break the check itself
We remove a protection on purpose and confirm the check turns red. A check that does not turn red is not counted as a check.
Weaknesses we find stay on a list until they are closed, with when and by whom. Current verification runs in a pre-release development environment.
What we do not cover
Honestly — what we cannot protect.
Letting you believe something is protected when it is not is worse than no protection at all. We work to soften these, but we do not guarantee them.
Large-scale flooding attacks
Attacks that flood a service with traffic from all over the world to knock it down. Limits at the door soften them, but we do not claim to stop them.
The physical and infrastructure layer
Break-ins at the data centre or on the machines themselves are covered by the hosting company.
Leaked sign-in details
We cannot prevent passwords or keys from leaking. We do provide ways to cancel them once a leak is noticed.
Unknown flaws in parts we use
Flaws nobody knows about yet in outside components we rely on. We check regularly, but the unknown is outside what we guarantee.
FAQ
What people ask.
No. There is no step in this system that trains AI on your data. What you bring in is used only to draw out its meaning and make it searchable.
No. Which company data belongs to is decided by the signed-in person’s membership; rewriting a request does not make you another company. Brands within one company are kept apart too.
Sending, publishing, money and deletion happen only with your approval, or within a scope, budget and time limit you set when delegating to the AI. Drafts and internal edits are made by the AI, and the previous version is kept. The AI cannot exceed the permissions of whoever asked, so it cannot do what they cannot.
Yes. People’s actions and the AI’s actions go into the same record. You can see who asked, what the AI did, and who approved it.
Yes. By role, and by limiting which brands each person can reach. By default, business numbers such as sales are for owners and admins, and who was paid how much is for owners only.
Sealed with strong encryption. Writing keys into program code is forbidden and checked for automatically.
There is a read-only connection, and connecting goes through a sign-in and a consent screen.
We think a vendor that never lists its limits is the riskier one. Knowing the exact scope is part of being safe. If anything concerns you, talk to us before signing up.
Before you hand anything over, ask us what concerns you.
We can also walk your IT team through the details. Please read our terms and privacy policy as well.