Aether
Design timeA threat model for AI and agentic systems. It draws the question before a line ships — who is allowed to do what, with which tool, on whose data — across architecture, agents, models, data paths, tools, and identity.
About Us
Vertotech is an AI research lab.
The lab studies how agents, models, and tool-using systems fail, then follows that into the system that has to hold: architecture, agents, models, data paths, tools, and identity.
A model can sound sure of itself and still be a careless colleague. Give it a goal and it will reach for a tool. It will keep going for as long as the system allows. The failure worth studying is rarely a clumsy sentence. It is a grant that should never have existed, a document the retriever trusted too easily, or an agent that did not stop when the business already had.
So the lab works backwards from the act. Who was allowed to do this, with which tool, on whose data, and what would have shown it in time? Aether draws that question at design time, as a threat model. Provenant presses the same question in the lab, against prompts, retrieval, tools, and grants.
The two meet in the middle. A threat model is only as good as the evidence that it held, and an evaluation is only as useful as the design decision it can change. We build both so the same question — what could act, and what would have caught it — is asked at the drawing board and again under load.
What belongs on this site is the shape of the problem. The procedures, the evidence, and the customer work stay inside the product.
What we build
Two views of one question
Provenant
In the labThe same question, pressed against a running system: prompts, retrieval, tools, and grants. It looks for the act that should not have been possible, and for what would have shown it in time.
The team
Experience behind the lab
The people behind Vertotech have spent years building, breaking, and defending production systems — from architecture and cloud infrastructure to identity, data, and the controls that hold under audit. That work now points at how AI and agentic systems fail, and what it takes to catch them before they act.
It is experience earned across several regulated, high-stakes industries — where a wrong action has real consequences and the evidence has to stand up afterwards.
Financial servicesHealthcareTechnology & SaaS
How we work
What the lab holds to
Work backwards from the act
The failure worth studying is rarely a clumsy sentence. It is a grant that should never have existed, or an agent that did not stop when the business already had. We start from the consequence and trace it back to the permission that allowed it.
Study the whole system, not the model
A model can sound sure of itself and still be a careless colleague. Give it a goal and it reaches for a tool, and keeps going for as long as the system allows. The controls that hold live in architecture, data paths, tools, and identity — not in the weights.
Publish the shape, keep the specifics
What belongs on this site is the shape of the problem. The procedures, the detector corpus, the evidence, and the customer work stay inside the product, where they can be held to account.
