Security · What we assume, and what is not done

State the boundary before calling the system secure.

Draft for review. This page describes the proposed design. Features without a confirmed implementation status are marked for review; they are not claims of running controls.

An agent uses a model to reason and a harness to act. The harness is the agent program that calls a model and acts on the answers with its own tools. A seat is the confined environment in which the harness runs. An administrator is a person with permission to change the system’s controls. A threat model is a statement of what the design protects, what can threaten it and what it assumes. The Awakener is the mechanism being built to start an agent within a limit and record what happened. A receipt records who asked, which agent ran and how it ended. Planted-fault testing introduces a deliberate defect to check whether a test or reviewer notices it.

The final topic in the outline is the threat model and its limits. This draft brings together the limits already stated in the Security section and the questions that still need an agreed answer.

It is a boundary statement for review, not a claim of a completed security assessment.

Design outline dated 6 October 2026. What the four status labels mean.

The proposed design

Each heading states the design or practice. Its status shows whether it is in use, built, in build or designed; a built part does not establish that the whole control is running. Where the outline gives no implementation status, that status awaits review.

  • Agents can be wrong or misled.

    The design starts with the possibility that an agent makes a mistake or treats something it reads as an instruction. Controls must sit outside the agent’s own willingness to comply.

    Implementation status awaits review.

  • Administrator authority is outside this protection boundary.

    The current design protects against agents and the programs around them. It does not protect against the administrator. A system with more than one administrator needs further design.

    Implementation status awaits review.

  • A reviewed part is not a completed control.

    The Awakener has reviewed parts of its record, while its complete path from permission through a bounded live run to a receipt remains unexercised.

    Implementation status awaits review.

  • Agent review is not an outside audit.

    Independent agent review and planted-fault testing are methods used by the team. They do not replace an outside audit, and no outside audit has occurred.

    Implementation status awaits review.

  • The rest of the threat model needs agreement.

    The protected assets, other threats, assumptions about the environment and the evidence needed for acceptance still need to be stated and reviewed. This draft does not assign them on the team’s behalf.

    Implementation status awaits review.

What this provides

The intended result is a reader who can distinguish the design’s stated boundary, reviewed parts and open questions before deciding what the system can be trusted to do.

What is not yet proven

No complete threat model or security assessment is established by the supplied outline. The assumptions outside the stated agent and administrator boundary remain for review.

No outside audit or complete live Awakener supervision path is claimed. The seven new topic pages are drafts; features with unconfirmed status are not operating guarantees.