Protection that will assess how an AI agent works through a task.
We are extending protection to the connections between an autonomous agent's steps and the scope of the task it was given. Qronos is in development; we will present availability and supported options once they are verified.
Individual steps can lead somewhere you did not intend.
An agent is asked to prepare an overview. It works through the source material, decides its own next steps and ends with a proposal of what to do. Each step on its own can be fine while the whole run leaves the remit it was given. What has to be assessed is therefore the connection between the steps and the boundary of the task, not a single step.
The three principles we are designing Qronos around.
Connections between steps
The aim is to assess a step in the context of the steps before it and of the task, not in isolation.
Assessment before the consequence
We propose that a step with a tangible consequence be assessed before it runs, not afterwards from the log.
An explainable record of the decision
We are preparing a record that shows what triggered a decision and which rule it followed.
What Corpilus protects today, and what Qronos is meant to add.
Corpilus provides specific checks in the interfaces it is connected to. Qronos is meant to extend that assessment to how an agent works through a task. The exact scope and the availability of Qronos are not part of the offering yet.
See the AI protection available todayIn the design we keep the decision separate from the execution: the part that carries a step out is not the part that decides whether it may run. Some directions, including quantum technologies, we treat as research — no current function or performance rests on them.
Qronos is in development. It is not part of the offering, it cannot be deployed today, and we are not naming a release date.