People
Choose which employees, service accounts, and roles can request governed work.
Security
You choose what Glaux can access and do.
Glaux puts company rules in the path of work. Admins decide which people and teams can use which knowledge, tools, models, and sensitive actions, and when a person must approve the next step.
Requested action
Send the summary outside the company workspace.
Before action
Rule applied: approval is required before anything is sent.
AfterwardResult recorded
The activity history records the request, decision, outcome, and policy context that applied.
Controls cover the people and the action.
Security starts with plain boundaries: who is asking, which team they are working in, what source or tool is involved, and whether the next step needs approval.
Teams
Group permissions by workspace or team so access follows how the company operates.
Tools
Make approved connections available only where they belong, with unsafe requests stopped before use.
Models
Set which model and provider choices are available for the work being done.
Sensitive actions
Require approval or stop work before high-impact actions such as external sharing or data export.
Approval and access requests
People can ask without leaving the task.
When work needs access to a source, tool, model, or sensitive action, the request stays attached to the reason for the work. Approvers see the context before deciding.
Access requested
Use customer-success knowledge source
Reason: answer a renewal-risk question with the approved source trail visible.
The user asks for a tool, source, model, or action from the task.The request keeps the reason, workspace, and approver path together.Work continues only when the request is approved or already allowed.
Prevention and evidence are different jobs.
The rule is checked before the action. The record is written after the result. Glaux should not make teams wait for after-the-fact evidence to find out whether an action was allowed.
Before action
Apply the rule
Check identity, workspace, approved source, tool, model, policy, and approval requirements before a governed read or side effect happens.
Afterward
Record the result
Keep the outcome visible through activity history, approval history, and audit-oriented admin views.
Activity history supports admin visibility.
Security and governance teams need to see where controls applied without turning internal secrets or raw configuration into visitor-facing explanations.
Recorded
Activity history
Admins can review what moved ahead, what needed approval, and what stopped before action.
Visible to admins
Admin visibility
Security reviewers get a governance view without exposing secrets, private prompts, or raw internal configuration.
Data and deployment diligence
Inventory first. Guarantees only when approved.
A security review should map the real categories involved: people and groups, approved knowledge sources, tool connections, model choices, skills, automations, approvals, access requests, and activity records.
What we confirm during diligence
Glaux treats deployment, data handling, and evidence sharing as review topics until the approved terms are clear for your environment.
Which deployment boundary applies to your environment?Which retention, residency, deletion, and export terms have been approved?Which security evidence can be shared with your reviewers?
For security teams
Review the controls behind governed work.
Authentication
Service calls are authenticated and versioned before protected Control Tower paths accept them.
Request integrity
Signed Control Tower service requests bind method, path and raw query, request ID, contract version, and body hash, with replay protection.
Isolation
Control Tower owns authorization, approvals, runtime activity, audit, and safe projections while clients can request and display decisions but cannot grant authority.
Redaction
Admin and activity surfaces should expose safe explanations and audit pivots, not secrets, private prompts, or raw credentials.
Audit
Audit-oriented admin views are designed for filtered lookup and controlled export rather than public log exposure.
Standards posture
Alignment language only.
Glaux security posture supports alignment with NIST AI RMF and OWASP agentic, MCP, and agentic skills guidance as design input. Standards are design input here, not proof of third-party review.