Security at Sherlock Agent

Sherlock Agent is an AI-powered investigation and remediation platform purpose-built for managed service providers and enterprise IT. Because our platform interacts with production endpoints, collects diagnostic telemetry, and can execute remediation actions, we hold ourselves to the highest standard of security engineering.

Our approach is built on five pillars: strong encryption everywhere, strict tenant isolation, comprehensive audit trails, safe and scoped remediation, and granular access controls. Each of these is detailed below.

Sherlock Agent is hosted on Microsoft Azure in Australia, leveraging Azure's enterprise-grade physical security, network controls, and compliance certifications. Our infrastructure is designed for high availability and resilience.

In a multi-tenant platform serving MSPs who each manage dozens of clients, tenant isolation is paramount. Sherlock Agent enforces strict boundaries so that no tenant can ever access another tenant's data, investigations, or endpoints.

• Logical data separation — every data record is scoped to a specific tenant. Cross-tenant queries are architecturally impossible at the data layer.
• Scoped access tokens — API tokens and session credentials are bound to a tenant context and cannot be used to access resources outside that scope.
• Data controls per tenant — MSP administrators can configure data retention, collection scope, and which diagnostic data is gathered for each client environment.
• Isolated execution contexts — investigation and remediation tasks are executed within isolated contexts to prevent cross-tenant interference.

Every action in Sherlock Agent produces an audit record. This is essential for MSPs who need to demonstrate compliance to their clients and for enterprise IT teams operating under regulatory frameworks.

• Full investigation audit trail — every investigation step is logged, including the data queried, the AI reasoning path, conclusions drawn, and actions recommended.
• Command logging — every command executed on an endpoint is recorded with the who, what, when, where, and the resulting output.
• Approval records — when remediation requires approval, the request, approver identity, timestamp, and decision are all captured immutably.
• Deployment tracking — every remediation deployment is logged with its scope, target endpoints, execution result, and rollback status.
• Exportable logs — audit data can be reviewed within the platform or exported for integration with your existing SIEM or compliance tooling.

Sherlock Agent can execute fixes, not just find problems. Because remediation touches production systems, we have built multiple layers of safety into every step of the process.

• Scoped execution — remediation scripts run with the minimum permissions required and are scoped to the specific endpoints and resources identified in the investigation.
• Approval workflows — critical or high-impact remediations require explicit human approval before execution. Approvers see the full remediation plan, affected systems, and risk assessment.
• Rollback plans — every remediation is generated with a corresponding rollback plan. If something goes wrong, changes can be reversed quickly and reliably.
• Dry run mode — remediation actions can be previewed in dry-run mode to verify their impact before committing to live execution, giving operators confidence before they press go.
• Blast radius controls — remediations can be deployed incrementally, starting with a single endpoint or a small group before rolling out to the full scope.

Sherlock Agent implements role-based access control with granular permissions, ensuring that every user sees only what they need and can do only what they are authorised to do.

• Role-based access control (RBAC) — predefined roles such as Administrator, Analyst, and Viewer provide sensible defaults, while custom roles allow fine-grained configuration.
• Granular permissions — permissions are scoped by action type (view, investigate, remediate, configure) and by resource (tenants, endpoint groups, investigation categories).
• Principle of least privilege — users and service accounts are granted the minimum access required for their function. Elevated access is time-bound and logged.
• Session management — sessions are secured with short-lived tokens, automatic expiry, and the ability to revoke sessions across all devices.