Stop customer data from leaking into AI — without banning it.
Guardrail catches PII, PHI, cards, and secrets on-device the moment they’re shared with ChatGPT, Gemini, or Copilot — then coaches or blocks. Deployed through your MSP across the endpoints they already manage. No bans. No disruption. Nothing sensitive leaves the machine.
Unsanctioned AI is the default now — and it’s invisible to everything you already pay for.
It looks exactly like legitimate traffic.
Sees a domain, not a secret
chatgpt.com over 443 is indistinguishable from any other site. The packet is encrypted; the payload is a patient record.
Watches files, not chat
Built to catch attachments and exfiltration. A pasted paragraph in a text box is not a file event — so it’s never inspected.
Allow or block — nothing between
Ban the domain and productivity revolts; allow it and you’re blind. Neither sees what is leaving. There is no middle setting — until now.
A seatbelt, not a roadblock.
Intercept
Guardrail catches the moment sensitive data is about to leave for an AI tool — before it ever sends.
Detect on-device
Sensitive patterns — PII, PHI, cards, secrets — are classified on the device in real time. Nothing sensitive is sent anywhere.
Coach or block
A quiet in-the-moment heads-up before it sends — or a hard block on the highest-risk data. The user keeps working.
Log & report
Every event feeds the dashboard — masked by default, audit-ready for compliance and cyber-insurance.
We never ban AI. Bans drive staff to phones and personal accounts — and owners want the productivity. Guardrail makes AI safe to use, which is why it survives in the real world.
One guardrail. Three reasons to deploy it.
Close the AI blind spot
See every sensitive egress, prove control for audits and cyber-insurance, and flip from monitor to block without a rollout. Detection runs on-device — so the evidence never becomes your liability.
A new recurring line on the invoice you already send
Force-push in one policy, near-zero support, ~2.5–3× managed-security margin — and it scales with your book. The free assessment is your demo, proposal, and trial in one.
A control that lowers the risk you underwrite
Give insureds a measurable AI data-leak control — with a live, exportable proof for cyber questionnaires and renewals. Differentiate the placement; lower the loss probability.
The questions you’re already asking.
It’s the AI layer your stack is missing.
Firewalls watch domains, EDR watches the endpoint, legacy DLP watches files — a pasted prompt is none of those, so it slips past all of them. Guardrail is defense-in-depth for AI egress: it sits beside the tools you already run and closes the one gap they share. It completes your defenses — it doesn’t replace them.
Monitor mode is invisible.
No prompts, no disruption — it runs under a posted policy. Protect mode is an in-the-moment heads-up, not a wall. A seatbelt. We never ban AI.
Detection runs on-device — we never see your content.
Classification happens locally, the instant before data would leave, so nothing sensitive is ever sent to us. Events are masked by default: you see the kind of thing that fired, not the secret itself. The evidence never becomes a new liability.
Don’t take our word for it. See your own exposure.
We deploy monitor-only for 14 days — no prompts, no disruption — then deliver a living report of exactly what almost left. The report is the demo, the proposal, and the trial, in one.
● Private beta — design-partner slots are limited