It started with a ban.
Like most stories about enterprise AI, ours begins with an IT team blocking ChatGPT, and everyone using it anyway.
- 01 · The ban era
Block it, and hope.
Every IT team we knew was answering AI the same way: ban it. It did not work. The tools were too useful, so people used them anyway, quietly, on their phones, with data nobody could see. IT pretended the ban held. Employees pretended they were not using AI. Nobody was actually in control.
policy: ban · status: bypassed
- 02 · We drew a line
So we built the thing we kept wishing existed.
A boundary between their AI and your data. Not surveillance. Not another ban. A control plane that decides, field by field, what a model is allowed to see, and never watches what your people do.
[Contact #7] · tokenised
- 03 · Today
Running in production, in the UK.
Boundary sits in front of real mailboxes, CRMs and ticket queues, making that decision on every request and writing every one of them down. We are a small team. We ship every week.
United Kingdom · in production
- 04 · What comes next
This becomes the standard.
We think deciding what models see becomes ordinary enterprise infrastructure, the way firewalls did. We are early, and that is deliberate: the teams that get AI control right now will set the standard everyone else follows.
the standard · forming
The founders, Boundary Control
We still answer every email. Talk to us
Six things we will not trade away.
We govern what the model sees. Never what your people do.
That distinction is the whole company. Boundary reads the data a request would expose, not the person making it, and it has no view of how anyone works.
Nobody should have to hand their assistant back.
Governance that starts by taking something away does not survive contact with the people it governs. Secure the data instead, and the tools stop being the argument.
When a rule is unclear, the answer is no.
Unclassifiable data does not move. That is a decision made once, in the product, so it cannot be quietly traded away later under delivery pressure.
We are infrastructure you control, not a party to your trust.
You set the tiers, you own the data, and you can revoke us. A control plane that its vendor can overrule is not a control plane.
Control you cannot audit is not control.
Every decision is recorded as evidence you can put in front of a regulator. If we cannot show our working, you have our word, and our word is not a control.
The safest data is the data we never keep.
Protecting what a model sees starts with not hoarding it ourselves. Resolvable detail expires on a rolling window; we are not a warehouse.
Put IT back in control.
See exactly what your AI can reach, and put a boundary around it.
Book an intro