Your AI agent acted. On what basis?
When an AI agent acts, you need to be able to tell later on what basis. A separate access key per agent and a log of every context retrieval make that provable.
An AI agent did something in your organisation: it answered a request, started a process, carried out a decision. Three weeks later someone asks about it. Which rules applied to the agent at that moment? In which role did it act? And which version of the policy did it know?
If you have no answer, you do not have a technical problem. You have a leadership problem.
Every agent gets its own key
In roleALPHA, an AI agent is part of the organisational model, with a purpose, a role and the policies that apply to it. The agent retrieves this frame as its context, and for that it gets its own access key. With this key it receives only its own context, never that of other agents.
Each key can be revoked on its own. When an agent is switched off or replaced, its access ends, and every other agent keeps working.
Every retrieval leaves a trace
Every time an agent’s context is delivered, roleALPHA writes an entry to the log, whichever way the request came in. The entry records:
- who retrieved it: the agent itself, a person or a connected AI application,
- which state of the context was delivered,
- which policies it contained, each with its version.
Someone who only looks at the agent in the app appears separately in the log, as a view. That keeps the record of what the agent actually received clean.
Traceability instead of a black box
For management, IT and HR this means: you can show which rules an agent knew at the moment it acted. That does not replace reviewing the individual case. But it answers the question every review starts with.
The agents keep running in your own environment. roleALPHA gives them the frame they are allowed to work in and records what they received from it.
What the EU AI Act has to do with it
The EU’s AI regulation (Regulation (EU) 2024/1689) requires, among other things, that high-risk AI systems log events automatically (Art. 12), that people can oversee them effectively (Art. 14) and that deployers keep the logs (Art. 26). A separate key per agent and a log of every context retrieval are one building block for this: proof of which rules an agent worked with. Which obligations apply to your use depends on the individual AI system.
Read more
- AI agents in the documentation: access keys and the retrieval log in detail
- Want to see what this looks like in your organisation? Request a demo