Security
Know what data can be accessed, where it goes, and who is responsible.
Each engagement documents the source systems, users, identity path, permitted tools, AI assistant provider, returned results, logging, monitoring, retention choices, incident responsibilities, and exit plan for the actual environment.
Access Controls
Start read-only and permit only the users, data, and operations the use case needs.
Read-only protects source records from change. It works alongside identity checks, purpose-built tools, result limits, and settings for the selected AI assistant.
Users and identity
Named users or roles and the authentication path available in the source and assistant environment.
Sources and fields
Approved interfaces, datasets, measures, dimensions, exclusions, and dates.
Permitted operations
Purpose-built read operations are preferred to unrestricted model-generated SQL or broad database authority.
Returned results
Results are shaped for the approved analysis and limited to what the user needs.
Visible failures
Denied access, unavailable sources, incomplete data, and unsupported questions should fail clearly.
Roles and Responsibilities
Systems Place operates the agreed service. The customer and providers keep their own authority.
The split is documented for the selected source, AI assistant, hosting, network, and identity design.
Managed service
- Implement the approved tools and service controls
- Maintain the answer tests and operating checks
- Monitor the agreed service and investigate its incidents
- Assess and test changes to the managed components
Business authority and product controls
- Approve users, sources, definitions, and business purposes
- Configure source and AI provider accounts and policies
- Own decisions made using the returned analysis
- Operate infrastructure and terms outside the managed service
Monitoring and Incidents
Check the service before launch and monitor it after.
The monitoring and incident plan is proportionate to the data, systems, provider, and business decision in scope.
01
Check answers
Compare representative answers with trusted reports, source results, and known examples.
02
Observe the service
Monitor approved tool requests, results, authentication failures, and service health.
03
Respond to incidents
Use named contacts, containment actions, investigation steps, and customer notifications.
04
Control changes
Retest relevant answers and access rules after material changes.
- Which events and failures are recorded
- Who receives and investigates an alert
- How access can be suspended or revoked
- When customers and providers must be involved
Common questions
Common security and data questions.
Does data stay inside our source system?
Not necessarily. The selected AI assistant provider may process the conversation, tool requests, and approved results. The exact data flow is documented for the chosen option.
Does read-only remove all risk?
No. Read-only prevents source-system writes, but sensitive results can still be returned, retained, copied, or shared. Identity, permissions, result limits, provider settings, and user behaviour still matter.
Do source-system permissions automatically apply?
It depends on the interface and identity design. Some options use end-user identity; others use a service identity and need tool-level enforcement. The design is verified for the actual source and assistant.
What is logged?
The agreed design may record service calls, tool selection, outcomes, authentication events, failures, and evaluation runs. Sensitive content and retention are deliberately scoped.
Review a real data path
Start with the question and the systems it would need to use.
The assessment records the source, provider, identity, tools, answer checks, retention, responsibilities, and exit requirements for the proposed option.