Forage Safety Principles
3 min read · Last updated · Page version 11
What is true today
- Forage deletes nothing. It edits no document in SharePoint or in the archive, and it never writes to Squirrel's record.
- A content label is a fact, not a decision. Recognising that a document contains payment card numbers says what it holds, and nothing more.
- Every content rule can be test run before it is switched on. The test run is offered rather than required.
The one change Forage can cause in storage
Where Squirrel has placed an archived file in Azure's offline archive tier, Forage asks Azure to bring it back online before reading it. Azure charges for this, it can take up to a day, and Forage does not return the file to the offline tier afterwards.
Delete them permanently
In Settings, Delete them permanently applies to a document type that has been switched off. It removes Forage's copy of the text, its search entries, and its record that those documents were examined. A reason is required and is recorded.
The documents in SharePoint and in the archive are untouched.
Privacy and isolation
- Only Full Administrators can reach it. The Records section is available only to the Squirrel portal's Full Administrators. Search results are not trimmed by SharePoint permissions, which is why access is kept to that role.
- No document content leaves the deployment. The text inside your files is never sent outside your own deployment. Where an administrator switches on emailed reports, the reports themselves go to the addresses chosen, and they carry figures, file names and site names, never the text inside a file.
- Classification is transparent rules. Every finding keeps the words that produced it, and the same document always gives the same result. No AI model decides which label a file gets. Scanned pages are read by optical character recognition that runs inside the deployment.
- AI that stays inside. Forage sends no document content to any external AI service. Its one AI model, Forage AI, runs inside your own deployment and only suggests labels, when an administrator asks for them. Every label is applied by a transparent rule that a person has switched on, and Forage does not train a model on your content.
- Text stays in the deployment. Forage keeps the text it reads, and its search index, inside your own deployment.
- One deployment per customer. Each customer has its own Forage deployment and its own index. There is no shared index, and no path for one customer's search to reach another customer's content.
See limitations for the full list of what is not yet available.
Questions about how a deployment is secured go to SmiKar (sales@smikar.com).