Safeguard / product direction
When something needs to pause, block or escalate, the response should happen locally. The record should remain available afterwards.
The line between control and record
AI systems need controls that can isolate, block, pause and revoke access. Orvessian does not take over those controls. It retains selected events and decisions they produce, so an authorised reviewer can understand an incident without relying only on the AI operator’s database.
Your security controls
Prevent and contain↓Orvessian trust record
Preserve and reviewThe intended safeguard flow
Record the selected policy version, authority and execution context that applied to a meaningful action.
Connect a selected tool request, monitor alert or enforcement result to the evidence retained by its custodian.
Show who approved, paused, blocked or revoked access, with the scope of that decision.
Give authorised reviewers a restricted, independently checkable record of what the surrounding controls observed.
Why independent anchoring matters
A shared ledger can give reviewers evidence that does not depend solely on the operator’s own records. The strength of that property depends on the real network, key management, source systems and verification process. It does not make the agent, sandbox or application immune to AI-assisted attack.
Read the technical boundaries →The boundary matters
It cannot guarantee complete detection, ensure source events were truthful or replace a runtime security control. It can make selected observed events, policy bindings and authorised interventions easier to inspect with their stated scope.
Availability
The agent-safety evidence profile and a generic policy-gateway example are next-phase development work. The current local receipt alpha does not offer real-time monitoring, automatic containment or a general safety verdict.