10 to the 23 AI logo

Practice Library

Governance patternauthority

Connection authorization

No data pathway exists until someone with authority approved it — the general edge-level access control.

What it changes

removedContaminated records repeated as fresh output(unauthorized pathways)
removedRecords replicated between systems(unauthorized replication)
cappedFailures relayed between AI models or agents(agent-to-agent couplings only where explicitly granted)
removedRecords replicated outside the governed system(record replication out of the boundary closed)

Who can pull it

Deploying organizationHarness builderData-protection officer

What it looks like institutionally

Systems accrete connections: a cache wired to an agent, a transcript fed to a fine-tune, a case system syncing to a analytics store. Each new pathway is a new route for error and a new surface nobody governs. Connection authorization inverts the default: pathways between components are closed until an accountable actor opens them, with the grant recorded.

This is ordinary security practice applied to error propagation: least-privilege for information flow, where the risk is both contamination inbound and exfiltration outbound — a pathway that copies records out of the governed system to an external store nobody holds a data agreement with is the same unauthorized connection, closed the same way.

The audit twin: periodically enumerate live pathways and reconcile against grants. The connections nobody remembers approving are precisely the ones to worry about.

Addresses: Unauthorized data pathways · Shadow integrations · Records replicated out of the boundary. Test a version of this lever in the PAN Lab.

Deciding whether this lever fits your deployment?

Which patterns matter, and in what order, depends on your system's actual shape. Ranking your options on evidence, with what can backfire stated, is engagement work.

Sources & Evidence

Claims made on this page and what supports them. The full registry lives in Evidence.

ScenarioIn the published runs, deleting records without reading them raised the contaminated share by stripping out be…

In the published runs, deleting records without reading them raised the contaminated share by stripping out benign entries; only content-aware cleanup reliably reduced it.

From the published runs: PAN governance-lever audit.