Discovery is not assurance
Observing an AI-related workload or identity creates a candidate for review. It does not prove that the asset is safe, compliant, or authorized to operate.
Ostelun’s product model is intentionally conservative. These principles describe the trust boundaries the platform is being built to maintain.
Observing an AI-related workload or identity creates a candidate for review. It does not prove that the asset is safe, compliant, or authorized to operate.
Bringing a candidate into the Agent Registry establishes an object to govern. The agent begins unassessed unless real evidence already supports a different state.
Where possible, Ostelun favors read-only access and bounded scope rather than broad tenant-wide permissions simply to make a workflow appear more automatic.
Machine collection, external ingestion, human review, and fixture/demo data are different trust statements. Ostelun is designed to keep those origins visible.
Strong identifiers can converge observations across sources. Names, owner hints, or fuzzy similarity do not automatically merge identities. Contradictory strong identifiers fail closed for review.
Discovery telemetry may describe what an agent can access. It must not silently become the authorization baseline for what that agent should be allowed to access.
For inbound security data, Ostelun distinguishes the time data was ingested from the time the upstream source says it observed the condition. Replaying old data should not make stale telemetry look current.
An Ostelun authorization decision is a governance state. It is not an enforcement action unless a customer explicitly connects an enforcement workflow.
Demo, fixture, development, and live/customer evidence should remain distinguishable. A polished demonstration is never a reason to blur provenance.
Ostelun’s strategy is to consume useful control-plane and security signals rather than become another endpoint sensor, inline gateway, or packet-inspection product.
We welcome architecture and trust-model scrutiny during design-partner conversations.