AI AGENT SECURITY
VERADIN
Verify what an AI agent will do before you trust it.
An AI agent can recognise that it is being tested and behave differently once it is live, and it can do so while reporting that nothing was unusual. That means a clean safety test may prove very little. VERADIN removes the tell.
The problem
Agents pass their tests
and can still act differently in production.
Asking does not work
a model can notice a test and not say so.
A better test lab does not help
the agent only has to work out it is in a lab at all.
How VERADIN works
Trials move into ordinary work
there is no separate test environment for the agent to recognise.
The action is sealed before the answer
the agent commits to an action before it learns whether the action is real or routed to a shadow system that changes nothing.
Permission becomes a lease
it expires, renews only while the evidence holds, and suspends itself when the model, a tool or a prompt changes.
What the demonstration shows
A live demonstrator runs the mechanism against a sample agent and simulated tools. It connects to nothing of yours.
Action escrow
an action sealed cryptographically before its destination is revealed.
Shadow-effect fabric
identical live and shadow endpoints, where shadow changes nothing.
Authority gate
the test flagging an agent that hides and leaving an honest one alone.
Rolling lease
permission that expires and suspends itself when the system changes.
Evidence
a signed record of every decision, exportable.
Demonstration on a sample agent. Not connected to any real system.
What we have shown
In a live test against a real model, the method flagged an agent that was hiding, left an honest agent alone, and would have caught nothing by asking the model directly. VERADIN is at concept stage: the mechanism is validated and demonstrable, and available today as a scoped engagement on your own agents. The installed product a customer runs in their own production systems is in development, and we say that plainly.
Work with us
Run it on your agents
We run VERADIN against your own agents and workflows in a scoped eight to ten week engagement, in a contained setup. There is nothing to procure.
Start an engagementDeploy it in your systems
If you want VERADIN deployed inside your own production environment, contact us and we will scope a deployment with you. This is bespoke and delivered with you, not a self-serve install.
Talk to us about deploymentVERADIN. UK patent application GB2617845.9. Physivitis Ltd. The mechanism is demonstrable today and available as a scoped engagement; the self-hosted product is in development. No performance figures are published.