It gates acts, not words. What an agent says about its own request is not evidence — and every refusal is on the record.
You gave an agent a job. You also gave it your keys.
To be useful, an agent needs access to your systems, your accounts, your customers. Once it has access, it can do things nobody approved. Not out of malice: because it inferred, or guessed.
You are accountable for what your agents do. Today, you have their word for it. An explanation is not evidence. An agent can describe what it did, and why, in fluent prose — and the description can be wrong, incomplete, or written after the fact. A board, a regulator or an insurer cannot act on it, and neither should you.
As AI reasoning becomes less human-readable, reading it stops being a form of oversight at all. What survives is oversight of what the agent actually did: which acts it attempted, which were allowed, which were refused, and on what evidence.
Your agents are not the risk. Unbounded authority is.
Aletheai Conduct sits between your agents and the systems they act on. The effect is three things.
Every act an agent takes passes through it, and the agent cannot reach it.
The act goes on the record first, so the record can never be a story written afterwards.
What an agent says about its own request is not evidence. The justification has to come from evidence the agent did not write.
What changes when your agents go through Aletheai Conduct, and why that is different from permissions, filters and logs.
An agent fully permitted to send that email is still held if nothing it can point to justifies it.
A record you can defend has to include the gaps. We built Aletheai Conduct on the view that a confident answer with nothing behind it is worse than no answer, and that the people who carry the accountability deserve to know which is which.
Who sign off on agents acting in the company's name, and need to be able to say what those agents did.
Who have to show a regulator the basis for every action, not a transcript of the reasoning.
Who price and certify exposure, and need a record they can re-derive themselves rather than take on trust.
We sell it to the people who are accountable for the agents, not to the people whose work it checks.
This page is a preview. The full description is shown under NDA.
Patents pending. ALETHEAI™ trade mark application filed.
Governing our own agent operations since October 2026.
None to name yet, and we will not invent any. We are looking for design partners who carry the accountability described above.
The mechanism, the live record and the confidential paper are shown under NDA, to the people who would be accountable for the agents. Tell us who you are and what your agents do.
Request access →