What is InterLock?
InterLock is open-source infrastructure for governing how AI agents access production data. It combines role- and policy-based access with redaction, approval for risky writes, and an audit record. It is intended for teams building AI systems that need to read or change real data, particularly when access must be controlled and agent activity needs to be traceable.
InterLock Features:
Production Data Access: InterLock provides governed access to production data for AI agents, supporting systems that need to work with real data.
Role and Policy: Access controls use roles and policies to define what an agent may read or change in the data it works with.
Data Redaction: Redaction is part of the access layer, helping teams control which data is exposed when agents work with production systems.
Risky Write Approval: Writes considered risky can require approval, adding a review step before an agent makes a consequential data change.
Audit Record: InterLock keeps a full audit record of agent access and actions, providing a trace of activity around production data.
InterLock Benefits:
Bounded Agent Access: Teams can bring agents closer to real data while keeping their permissions governed, rather than granting unrestricted access by default.
Reduced Data Exposure: Redaction helps limit the information available to an agent, reducing the chance that its work requires exposure to unnecessary data.
More Deliberate Changes: Approval for risky writes gives teams a chance to review sensitive changes before they are carried out by an agent.
Traceable Activity: An audit record makes agent activity easier to follow, supporting teams that need to understand what happened around their data.
Use Cases:
Production Data Queries: Use InterLock when an agent needs to read production data and the team wants access governed by roles and policies.
Agent Data Updates: Apply it to agent workflows that change production data, especially when some writes should be held for approval.
Controlled Data Exposure: Use redaction in workflows where agents need to work with production data but should not see every available detail.
Action Investigation: Consult the audit record when a team needs to trace an agent's activity involving production data or understand what it changed.