Identity and delegation
Every consequential action identifies the principal, role, purpose, scope and expiry.
Security and Human Control
Bounded operational responsibility, never unrestricted autonomy.
Coda operates inside explicit customer-approved authority, coverage and service boundaries. High-impact, irreversible or insufficiently observable actions remain approval-gated or prohibited.
Control model
Six boundaries protect the operating relationship.
Action boundaries
Actions are classified as observable, approval-gated, bounded autonomous or prohibited.
Human sponsorship
A named customer authority approves the mandate and retains veto over consequential changes.
Isolation
Operating credentials, environments and tools are separated by guild, customer and purpose.
Rollback and compensation
Autonomous remediation is limited to actions that are reversible or explicitly compensatable.
Evidence preservation
Decision, authority, action and outcome records remain available for inspection and replay.
Graceful degradation
When controls weaken, autonomy contracts.
Bounded autonomy
Approved actions execute inside the operating contract.
Read-only mode
Observe and investigate without production mutation.
Human-supervised mode
Every consequential action requires named approval.
No execution
Revoke authority, preserve evidence and initiate incident response.
Customer control
Coda responsibility
Approve mandate and service boundary
Customer defines the business and risk envelope.
Operate within the approved envelope
Coda does not expand its own authority.
Name sponsors and approvers
Human accountability stays explicit.
Preserve escalation and evidence
Coda presents decisions with inspectable context.
Control data residency and access
Customer requirements shape deployment architecture.
Apply least privilege and isolation
Access remains scoped to the operating purpose.