PRODUCT · DATA LEAK PREVENTION

CATCHES PERSONAL DATA IN AND OUT.

Your AI does not have to be hacked to leak. It just has to be helpful at the wrong moment. A support agent pastes a record into a chatbot. A workflow ships a file to a model that trains on it. Mountain Theory checks the action, not the intent: data that is allowed to move, moves. Data that is not, stops before it leaves.

What it does

Identity controls decide who can open a record. Nothing decides where the AI takes it next. The check here is on the action, so it holds whether the data was typed, pasted, or pulled from a system the agent was allowed to read, and whether it is on the way into a model or on the way out. Every action gets one of three outcomes, ALLOW, HOLD, or BLOCK, and every decision is logged. Built for the places where one leaked record is an immediate violation: FERPA in schools and universities, HIPAA in hospitals and clinics, PCI in payment systems, GDPR across the EU, SOX in financial reporting.

The scenario

A vendor updated the hosted model behind one of our own agents overnight, and the agent began reaching for the network to answer questions it used to answer from local records, improvising commands instead of the sanctioned scripts. Every attempt was stopped the day it appeared, with no new rule on our side. In the ungoverned lane of the same run, every one of them completed.

The July run, recorded, with the two actions we had no policy for

One rule, as you would write it

Policy

Personal data, health records and payment data do not leave the environment in any AI action, in or out. No approval path. Log every attempt.

Plain English, written by whoever owns the risk, no code to change it. Add a HOLD for the one transfer a person should see, a record leaving for a vendor, and everything else runs at full speed.

What the record has to contain, and why the usual logs do not count

The ask

A 30-day paid proof of concept on your agents and your action set, in your environment, not ours. Then one flat annual price for the organization, with no per-agent count. Tests published, misses included, before anyone signs.

Book 30 minutes and bring the workflow that touches the records

Every published test, on one page →

Proof

Four things with a published run behind them.

Tool chainingBlocked once, blocked again on the workaround.The Optimo AI run →Agent driftThe goal it was given, not the one it wandered to.The July run →Zero-day behaviorStopped the day it appeared.The July run →No new ruleNo signature, no patch. Nothing on our side changed.What we did not have to do →

Tests published, misses included. The Recorded Run Protocol →

Where next

See the recorded runs →Compare all 77 vendors →How it works →Read the FAQ →Contact us for more info →