Security at Snowfence: what it sees and what it keeps
Snowfence checks data where it already is, on your own laptops and servers, in your own infrastructure. Your prompts, files and clipboard never reach Snowfence. The console your security team uses keeps metadata only: which rule matched, which label, which device, and when.
Where do the checks run?
Every check that looks at content runs on the laptop or server where the data already is, in your own environment. Prompts, files and tool results are judged there and never copied anywhere else.
Snowfence reads the content of a request only when it is going to an AI model service, so it can check the prompt and the model’s reply. Every other connection is judged by where it is going, without reading what is in it.
What does the console keep?
The console is where your security team writes rules, sees events and reviews cases. It holds:
- rule IDs, data labels, device and process names, and times;
- the reason a person gave when they chose to send something after being asked;
- signed statements from verified calls.
It never holds prompts, files, clipboard text, meeting audio or transcripts. Where a check needs to note which secret matched, it stores a keyed fingerprint. The key stays with you, so the console can’t work back to the value.
Each customer’s events and keys are stored apart from every other customer’s.
Can the console change the rules on its own?
No. Each laptop and server accepts only rules signed with your company’s own keys. Loosening a rule, such as moving data to a less strict tier, needs two people to sign it. The console can’t loosen a rule, and it can’t run anything on a laptop or server.
Does Snowfence watch how people work?
No. Snowfence doesn’t record keystrokes, screen contents, browsing history or working hours, and it doesn’t score people.
It does record attempts to get round its own rules, such as trying the same blocked file again in another browser. Those attempts are grouped into a case for your security team to review. Every block tells the person that attempts are recorded, cases show a pseudonym by default, and records are deleted after a set time. People decide what a case means, not Snowfence.
What Snowfence never does to your data
Snowfence decides; it never changes data. It allows, asks or blocks. It never rewrites a prompt, removes part of a file, or replaces the clipboard. A blocked tool call is cancelled, not altered.
What it can’t stop
Being clear about limits is part of security.
- A photo of the screen, or a second device. No signal reaches Snowfence.
- Personal devices. Snowfence runs on company laptops and servers.
- A local administrator. No software on a laptop can stop its own administrator. Snowfence records it, and that laptop can no longer sign in to company AI services.
- Malware that already controls the machine. Snowfence is not antivirus.
- Agents running in a vendor’s cloud. Snowfence controls who signs in, not what those agents do on the vendor’s computers.
Reporting a security problem
If you think you’ve found a security problem in Snowfence or this website, email pratik@snowfence.ai and say it is a security report. We’ll reply within a day.
Related
Updated