SARVA System for Aligned Runtime Verification & Authorization
SARVA evaluates supported requests against deterministic governance controls, authority context and organizational policy, and resolves each to one of three outcomes: allow, escalate or block.
Capability
what it can do
≠Authority
what it may invoke
≠Permission
what policy allows
An AI may be technically capable of an action without being authorized to perform it — and authorization is not the same as being permitted by policy. SARVA keeps an AI’s proposal structurally separate from execution authority.
If a gate determines a supported request should not proceed, the pipeline resolves it as escalated or blocked, and the decision is recorded.
Evaluates the request against the ethical and safety boundaries defined in policy.
Has this agent been authorized to perform this type of action?
Is this a recognized action type? No prohibited payloads.
Does active policy and role permit this specific request right now?
Addresses the authority boundary before supported consequential external effects.
Governed decisions are recorded as integrity-protected evidence in COSMOS, linked to the policy version in force.
Every supported request resolves to one of three outcomes.
The request is permitted on the supported path. The decision is recorded.
The request requires further evaluation or human review before proceeding. It may resolve to allow or block.
The request does not meet policy and is not permitted on the supported path. The decision is recorded.
Addresses the authority boundary before supported consequential external effects.
Authority is not assumed to last indefinitely. It is re-confirmed rather than carried forward.
Supported governed requests are evaluated against the current applicable policy, preventing earlier policy state from being treated as current authority.
Authority does not automatically pass from one agent or step to the next.
Designed to prevent authority from silently expanding as requests move through supported agents, services and workflows.
Governed decisions are recorded with context and the policy version in force. Supported requests are evaluated before they proceed on the supported path.
Governed decisions are recorded with context, timestamps and policy references.
Each governed decision is linked to the policy version in force at the time.
Supported requests that fail policy are resolved as blocked and recorded.
Evidence lives in COSMOS. Explore COSMOS
The core governance and evidence architecture is implemented and tested. MRCortex is completing the enforced consequential-execution architecture required for Customer-Ready V1.
Customer-Ready V1 in development
Identifies AI systems and their access context. Explore Discovery
Evaluates supported requests and returns a decision. (this page)
Records governed decisions as integrity-protected evidence. Explore COSMOS