FAQ
The questions buyers ask, answered straight.
Every answer leads with the answer, so you can skim. We start with what Verillian is and covers, then the ChatGPT and HIPAA questions most visitors bring, what running it feels like, and how the pilot works, which is a conversation we keep deliberately easy to start.
01
How Verillian works and what it covers
Verillian is AI governance for regulated work: a layer on your own devices that decides what any AI tool or agent may do, screens sensitive values on the way out, and seals every captured interaction into a record you can hand an auditor. Nothing of yours reaches us.
It sits on the device, in the path between the AI tool and everything that tool can reach, where a request is still readable. Every prompt and tool call is checked against policy you declared, then allowed, refused, or redacted in that moment, so the ruling lands before the action runs rather than in a report the next morning.
No. Your people keep the AI tools they already use, exactly as they use them, with nothing to install in the tool and nothing new to learn. The layer only makes itself felt when someone tries something outside policy, which is precisely when you'd want it felt.
Any AI tool that reaches a governed provider from the device: chat assistants, coding agents, command-line tools, and the shadow AI nobody approved. You govern providers rather than integrating tools, so a brand-new app needs no setup and no vendor's cooperation, and the trust page says exactly where enforcement stands provider by provider.
Those controls live downstream of the device and speak in traffic and files, so an AI action mostly reaches them as a request to inspect rather than a tool call to rule on. Verillian sits where the request is still fully readable, which is why it can refuse the action itself before it runs and sign a record of what did.
Most AI gateways put a model in front of the model: one AI judges what another is about to do, from a vendor's cloud. That judge can be argued out of its own instructions, and it only sees what you agree to route through it. Verillian's decision is a deterministic policy check made on the device, so the tools nobody approved answer to it too.
Yes. One control halts AI fleet-wide in a single move, and because the stop lands on the connection itself, the tools nobody approved go quiet with everything else. The record keeps who flipped it and when. When the problem is narrower, scope a stop to the user or device involved, and an autonomous agent running there is refused before its next action without stopping everything.
02
ChatGPT, HIPAA, and staying compliant
No, not on its own. ChatGPT doesn't become HIPAA compliant by policy memo, and pasting patient details into an ungoverned chat leaves you no control and no record. Governed use is different: identifiers screened before a prompt leaves the device, and every captured interaction sealed into an audit trail you hold.
It can be defensible, once your organization approves it. HIPAA turns on the specific service, its configuration, and any business associate agreement it calls for, and that approval stays with your compliance program. What Verillian adds is the technical half: identifiers screened on the device, policy deciding what the tool may do, and a sealed record that shows an auditor what happened.
AI policy enforcement means rules applied while AI acts, not reviewed after it's done. Verillian checks every captured request and tool call on the device against policy you declared, allows or refuses it in the moment, and seals the verdict, so the decision is made before the action runs, not after.
Security as control means the decision comes before the action: a system decides what an AI may do at the moment it acts, where detection can only report what it already did. Verillian sits on the control side, enforcing declared policy on the device and sealing the verdict into the record.
Shadow AI is AI used without approval: personal accounts, browser chats, tools nobody vetted. And yes, it can be governed rather than just discovered. Because Verillian sits on the device, policy binds those tools the moment they act, no cooperation from the tool required.
An AI audit trail is a record of what your AI was asked, what it did, and what ruling it met. Verillian's is tamper-evident: entries are signed and chained so any later edit shows, which is how a legal team answers from evidence months after the fact.
03
Production and deployment
About as long as a normal software push. The checkpoint rolls out through the MDM you already run, whether Jamf, Intune, or Group Policy. The fleet enrolls in the same window, and there's no per-tool integration to build first. The slow part of most rollouts just isn't there.
You author a rule once and scope it by role, and it reaches the machines it applies to. A clinical team, an engineering team, and a contractor can each get a different answer to the same question. Exceptions can be granted and scoped, and the record keeps which version of policy was in force for any decision.
It signs in through your own IdP over OIDC, Okta, Entra ID, or Google Workspace among them, so there are no local passwords to manage. Your directory's groups map onto console roles, the permissions you already run carry over, and MFA stays whatever your IdP enforces.
Very little by design, and you shouldn't take our word for it. The ruling is a deterministic policy check made right on the device: no classifier to run, no cloud round trip, nothing extra to wait on. The pilot is the honest benchmark, run on your own fleet under your own workloads.
Governance keeps working. Policy is enforced on the device itself, so no connection is needed to decide, and captured records queue locally until the link returns, then upload. If the local store ever fills, AI stops rather than running unrecorded, because fail-closed is the standing rule, and there's no setting that turns it off.
Mac, Windows, and Linux today, deployed like the rest of your endpoint software. The admin server it reports to runs wherever you choose, on-prem, private cloud, or fully air-gapped, so the deployment fits the boundary your data already has to live inside.
04
Proof, privacy, and your data
A sealed export of the records in scope, plus the means to check them. An auditor can confirm the chain is unbroken and every entry came from your devices without being shown your content, and if you grant content access, each entry they open is recorded too.
Because a change breaks it visibly. Entries are hash-chained and signed by the device that wrote them, and a single edited field shows up as a break anyone can verify. The record earns belief by being checkable, which is what makes it evidence rather than a log.
On servers you run, encrypted under a key only you hold. There's no Verillian-hosted copy of any of it: we never receive your prompts, responses, or records, encrypted or otherwise. The admin server is yours, and it stores the sealed content under your key alongside the metadata your team searches on. The reviewers you authorize can open entries, each open is itself written down, and no single administrator can grant content access to themselves.
No, and structurally it can't be. Training on your data would require having it, and nothing of yours ever reaches us: not prompts, not responses, not the record. As for the AI providers themselves, your policy decides which of them are reachable in the first place.
05
Pricing, the pilot, and licensing
It's priced per device, per year, as a subscription. Pricing in regulated sectors is a conversation about fleet size and term rather than a public rate card, so tell us your device count and we'll put a number on it.
It's a full deployment of Verillian in your environment: your endpoints, your policies, your keys, scoped to the rules you answer to. The records it seals are yours to keep whether or not you buy. The terms are agreed in the conversation, and the scope is what decides. Book the demo and ask.
Through a signed key, checked offline. Verillian is proprietary software, licensed per deployment: the key sets your device count and term, only we can sign a valid one, and there is no remote switch on our side. Your systems stay yours, license enforcement included.
Yes, without a new negotiation. Tell us the new count and we'll issue an updated key that runs through your existing renewal date, with the added devices prorated for the remainder of the term. Reductions happen at renewal, so scaling up never waits on a contract cycle.
Your devices keep enforcing the policy you last signed, and the record keeps writing. What pauses is change: policy edits and new device enrollment wait until the license renews. Everything sealed stays in your environment under your key either way, because your data was never with us, and any grace terms are agreed in your contract.
Still holding a question?
Ask it in the demo, where the answer can be a demonstration instead of a paragraph. And if you came for the fine print, the trust center is where we state it.